# GigHz Precision AI — full documentation (v13.35.3, docs v1.2.2) Each section below is one page from https://precision.gighz.com/docs/. Content is plain Markdown; tokens are already resolved. --- # What Precision AI is URL: https://precision.gighz.com/docs/overview/ Last reviewed: 2026-10-13 GigHz Precision AI is a web application for radiologists that turns a natural, spoken description of a study into a complete, structured radiology report. You name the study, describe the findings in whatever order you notice them, and the report assembles itself, section by section, as you speak. When you stop, the impression is composed from what you dictated, ordered by clinical priority, and the whole report is ready to review, edit and sign. It is built by GigHz LLC and used by attending radiologists, fellows and residents. This page is the orientation; the rest of the documentation goes section by section. ## The problem it solves Conventional dictation produces prose that someone, usually the radiologist, has to reorganise into the structured format most departments require. Templated reporting solves the structure problem but makes the radiologist navigate fields, which breaks concentration on the images. GigHz Precision AI keeps the radiologist looking at the images and speaking naturally, while the structure is handled by the software. ## How it works, in one paragraph Speech is transcribed live. Each sentence is handed to a placement engine that understands the template for the named study and files the finding on the correct line, replacing the template's default normal for that organ. Findings that fit no line are kept under an "Additional findings" heading so nothing dictated is ever lost. When dictation stops, an impression composer reads the dictated findings and the clinical indication, writes a numbered impression that answers the clinical question first, and a deterministic ordering layer makes sure emergencies always lead and incidental findings never do. The report is then a draft for the radiologist to review. > Every report is a draft. GigHz Precision AI is clinical decision support for the reading radiologist; it does not make a diagnosis and nothing it produces is final until the radiologist has reviewed and signed it. ## What is in the box | Capability | What it does | Documentation | |---|---|---| | Live report fill | Findings land on the correct template lines as you dictate | [Live report fill](/docs/live-report-fill/) | | Structured templates | ACR-style templates for CT, MRI, ultrasound, radiography, fluoroscopy, nuclear medicine, breast imaging and interventional procedures; each can be made your own | [Templates](/docs/templates/) | | Impression composition | A numbered impression that answers the indication first, ordered by clinical priority, with incidental findings kept out | [Impression policy](/docs/impression-policy/) | | Clinical decision support | LI-RADS, Bosniak, Fleischner, TI-RADS, PI-RADS, Lung-RADS, O-RADS and PECARN applied from the dictated findings | [Clinical decision support](/docs/clinical-decision-support/) | | Voice commands | Editing, undo, template switching and report control by voice | [Voice commands](/docs/voice-commands/) | | Output options | Patient-education paragraph, billing suggestions, signature block | [Output options](/docs/output-options/) | | Reference tools | Imaging protocol lookup and ACR Appropriateness Criteria lookup inside the app | [Reference tools](/docs/reference-tools/) | ## Who builds it GigHz Precision AI is built by GigHz LLC, a California company founded by a practising radiologist. GigHz is a member of [NVIDIA Inception](https://www.nvidia.com/en-us/startups/), NVIDIA's program for AI startups. The parent site is [gighz.com](https://gighz.com/). ## Who it is for Practicing radiologists who want structured reports without field-by-field navigation; residents and fellows, who get a trainee price; and groups evaluating AI-assisted reporting. It runs in a web browser with a microphone and needs no installation, no PACS integration and no IT project. ## What it is not It is not a PACS, a RIS or a dictation system that integrates with your worklist; reports are copied into your reporting system. It is not an image-reading AI: the product works from what you dictate, not from the pixels. And it is not a substitute for the radiologist's judgment, which is why the Terms require review of every report. ## Where to go next - New user: [Getting started](/docs/getting-started/) and the illustrated [Getting started guide](/guide/). - Evaluating for a department: [HIPAA, BAAs and PHI](/docs/hipaa-and-baa/), [Security architecture](/docs/security-architecture/), [Data handling and retention](/docs/data-handling/). - Pricing: [Plans, trials and billing](/docs/pricing-and-trials/). --- # Getting started URL: https://precision.gighz.com/docs/getting-started/ Last reviewed: 2026-10-20 This page walks through a first report from a fresh account. The illustrated version, with pictures of every screen, is the [Getting started guide](/guide/). ## Before you begin - A desktop or laptop browser. Chrome and Edge are the best tested; the microphone must be allowed for https://precision.gighz.com/. - A microphone. A headset or a dictation microphone gives the transcriber the cleanest audio; a laptop microphone works in a quiet room. - Your account. New accounts are created at [https://precision.gighz.com/signup/](https://precision.gighz.com/signup/); the password-setting link arrives by email. ## See a whole report built in ninety seconds A CT abdomen and pelvis dictated the way you would say it at the workstation, with the dictation box visible while the report fills in. More recordings are on the [Demo videos](/docs/demo-videos/) page. @video R3seew6SJUQ ## The basic workflow 1. Open the application at [https://precision.gighz.com/app/](https://precision.gighz.com/app/) and click the green **Start Dictating** button (or press F4). 2. **Name the study first.** "CT abdomen and pelvis with contrast", "MRI of the lumbar spine without contrast", "chest X-ray, two views". The template for that study appears on screen within a second or two. 3. **Describe the findings in any order.** You do not need to follow the template's order and you do not need to dictate normals; every organ you do not mention keeps the template's standard normal line. Measurements, laterality and image references are kept exactly as you say them. 4. Watch the report fill. Each sentence is placed on its line as you speak, and a counter shows how many sentences are still being placed. A finding that fits no line is kept under *Additional findings* with a flag, so nothing is dropped. 5. Click **Stop**. Give it a moment to place the last sentences and compose the impression. 6. Review. Click any line to hand-edit it. If you would rather change the transcript, edit the text and click **Rebuild**, which places everything again from the edited transcript. 7. Click **Finalize Report**, then **Copy to Clipboard**, and paste into your reporting system. ## An example you can say out loud > CT abdomen and pelvis with contrast. Right lower quadrant pain. The appendix is dilated to 11 millimeters with wall thickening and periappendiceal fat stranding. No abscess. No free air. Mild reactive right lower quadrant lymph nodes. Liver, spleen, pancreas, kidneys and adrenal glands are unremarkable. That dictation produces a complete report with the appendix finding on the gastrointestinal line, the nodes on the lymph node line, the normals you stated and the normals you did not, and a one-line impression of acute uncomplicated appendicitis. ## Punctuation and sentence breaks Say "period" or "next line" when you want a sentence or line to end. The transcriber also inserts its own sentence breaks when you pause, and those are not trusted: a pause in the middle of "7 millimeter right hepatic lobe cyst" does not split the finding. The placement engine re-joins fragments and decides where findings begin and end from the words, not from the pauses. ## Typing instead of speaking Everything above works with typed text too. Type or paste the findings and press **Create Report**. The same templates, the same impression rules and the same clinical scoring apply. ## If the wrong template appears The detected study is shown as a chip above the transcript. Click it to choose a different template, or just keep dictating: naming the study more precisely ("CT urogram" rather than "CT abdomen") switches the template and everything already dictated is re-placed into the new one. ## Common first-session questions - **Why did a sentence land under Additional findings?** The engine could not match it to a line with confidence. The sentence is kept verbatim with a review flag; click it to move or edit it. - **The impression says less than I expected.** By design. Simple atelectasis, unchanged degenerative change, simple cysts, vascular calcification and similar incidental findings are kept out of the impression unless something about them is urgent or changed. See [Impression policy](/docs/impression-policy/). - **Can I change the template's wording?** Yes. Click **Make it mine** above the transcript. See [Templates](/docs/templates/). ## Getting help Use **Report an issue with this dictation** at the bottom of any report, or email [signal@gighz.com](mailto:signal@gighz.com). Issue reports send metadata about the session, never the text of your report. --- # Demo videos URL: https://precision.gighz.com/docs/demo-videos/ Last reviewed: 2026-10-20 These recordings show GigHz Precision AI doing exactly what the rest of this documentation describes, on the live application, with synthetic dictations. No patient data appears in any of them. Each one is under two minutes. The dictation box floats at the bottom right of the frame so you can watch the words arrive and watch them land on the report at the same time. Every video is a click-to-play thumbnail. Nothing is sent to YouTube until you press play, which matters if you are reading this from a workstation with restricted network policies. ## The full workflow in four and a half minutes Start here if you have not seen the application before. Several cases are dictated naturally from beginning to end: naming the study, describing the findings in any order, watching the live report fill in, confirming the scoring that was derived from the findings, and finalizing and copying the finished report. It was recorded on an earlier release, so a few panels look slightly different from today's, but every step is the same. @video YBteTnWemds ## Dictate naturally, the report builds itself A CT abdomen and pelvis for right lower quadrant pain, dictated one thought per line, the way a radiologist actually talks at the workstation. The study sentence supplies the indication without the word "indication" ever being said. Each finding lands on its organ line as it is spoken, a sentence such as "solid organs are unremarkable" keeps every solid-organ line at its standard normal, and the impression is written from the findings in order of clinical priority. The diverticulosis stays in the findings because it does not answer the clinical question. @video R3seew6SJUQ ## Fleischner criteria from the nodule you describe A follow-up CT for a solid 7 mm right upper lobe nodule in a never smoker. Size, solidity, number of nodules and the risk category are read from the dictation. The Fleischner 2017 recommendation appears in a panel beside the report for confirmation, then enters the impression as its own numbered item. The image and series citation stays in the findings, where it belongs. @video 2oK3T3asdH4 ## TI-RADS scored while you speak A thyroid ultrasound with a 1.8 cm solid hypoechoic nodule described in one line. The five ACR TI-RADS feature categories are scored from that description, the panel shows the points per feature and the resulting level, and the size threshold for fine-needle aspiration is applied from the published table. A feature that was not described is shown as not stated and is not scored. @video uaPC9XEtAEQ ## LI-RADS derived from the dictation A surveillance MRI in a cirrhotic liver. The application knows the liver is at risk because the dictation says cirrhotic and HCC surveillance, parses the observation size and the major features from one sentence, and derives the LI-RADS v2018 category. The radiologist confirms it and it enters the impression in canonical wording, with the dictated description of the observation preserved. @video D7MZogmzjEs ## A complete CT abdomen and pelvis, ten sentences A CT for left flank pain dictated start to finish: an obstructing distal ureteral stone with mild hydroureteronephrosis, a hyperdense renal lesion whose density and enhancement are scored as Bosniak II and confirmed, hepatic steatosis, sigmoid diverticulosis, a normal appendix, a small umbilical hernia. Every sentence lands on its organ line, the organs not mentioned keep their standard normals, and the impression leads with the stone, then the cyst, and leaves the incidentals in the findings. @video 6mhKx8L6fQk ## A CT pulmonary angiogram with right heart strain The report every radiologist writes at two in the morning: bilateral segmental and subsegmental emboli, an RV to LV ratio of 1.2 with septal flattening, a small effusion, a wedge-shaped opacity consistent with infarct, and the incidental 6 mm nodule that is easy to lose in an emergency read. The nodule triggers the Fleischner 2017 panel for a high-risk patient, and the follow-up recommendation enters the impression as its own numbered item. @video GAQD3VO7ov0 ## An MRI lumbar spine, level by level Low back pain with left leg radiculopathy. Alignment, the conus, and each level are dictated as they would be said at the workstation: an L4-5 left paracentral extrusion with caudal migration contacting and displacing the descending L5 nerve root, an L5-S1 bulge with facet arthropathy, moderate bilateral foraminal narrowing and Modic type 1 changes, a mild L3-4 bulge, and an incidental renal cyst. The impression leads with the level that explains the symptoms. @video Qzsi0HeCmis ## What the videos do not show They do not show image analysis, because GigHz Precision AI does not analyse images. They do not show a PACS or RIS integration, because the product is a web application whose output you copy or export. And they are demonstrations, not validation: every report in them is a draft for the reading radiologist, exactly as it is in daily use. See [Clinical decision support](/docs/clinical-decision-support/) for how each scoring system is triggered and confirmed, and [Getting started](/docs/getting-started/) to try the same dictations yourself. --- # Evaluating for a department or group URL: https://precision.gighz.com/docs/for-departments/ Last reviewed: 2026-10-13 Most evaluations cover the same ground. This page lists it in the order it usually comes up, with a link to the page that holds the detail, so an evaluation can be done against the documentation rather than against a sales call. ## Clinical fit - Which studies are covered: [Templates](/docs/templates/). The library covers the common CT, MRI, ultrasound, radiography, fluoroscopy, nuclear medicine, breast and interventional examinations, with 107 studies mapped to live-fill bases, and any template can be adapted to a group's house style. - How the impression is composed and what it leaves out: [Impression policy](/docs/impression-policy/). Groups with a different convention can change the ordering and suppression behaviour per template. - Which scoring systems are applied and how they are confirmed: [Clinical decision support](/docs/clinical-decision-support/). - The reviewer's role: [AI models and limitations](/docs/ai-models-and-limitations/). Every report is a draft; the Terms require review. ## Privacy, security and compliance - Identifier policy and agreements: [HIPAA, BAAs and PHI](/docs/hipaa-and-baa/). The product does not need PHI and the Terms prohibit entering it; a BAA with GigHz LLC is available for covered entities that intend to use identifiable data. - Hosting, transport, scripts, roles, secrets, email: [Security architecture](/docs/security-architecture/). - Inventory of what is stored and for how long, and deletion: [Data handling and retention](/docs/data-handling/). - Testing and change control: [Quality assurance](/docs/quality-assurance/). ## Operational fit - No installation, no PACS or RIS integration, no IT project: it runs in a browser with a microphone beside the PACS. Reports are copied into the reporting system. - Each radiologist has an individual account; there are no shared logins. - Browser: Chrome or Edge on a desktop, microphone permitted for the application domain. Hospital browser policies that block microphone access block the product. ## Commercial - Two monthly prices and two trials, no annual plan, no lock-in: [Plans, trials and billing](/docs/pricing-and-trials/). - Group pricing and agreements are handled directly: [signal@gighz.com](mailto:signal@gighz.com). ## How a pilot usually runs 1. Two or three radiologists create accounts with their institutional email; the 90-day trial needs no card and no procurement. 2. Each dictates a normal day's work for two weeks, copying reports into the reporting system as usual. Nothing about the department's systems changes. 3. Feedback goes through the in-app issue report (metadata only) or email; fixes ship continuously and each becomes a permanent test. 4. The group decides on the evidence of its own reports, and the templates it adapted during the pilot stay with the accounts. ## Questions this page does not answer Email [signal@gighz.com](mailto:signal@gighz.com). Security questionnaires are answered from the documentation above; if a question is not covered here, the answer is added to these pages so the next group does not have to ask. --- # Live report fill URL: https://precision.gighz.com/docs/live-report-fill/ Last reviewed: 2026-10-13 Live report fill is the core of GigHz Precision AI: the report builds itself while you dictate, so by the time you stop talking most of the work is done. This page describes the behaviour you will see and the reasoning behind it. ## The skeleton appears when you name the study As soon as the transcript names a study, the matching template appears as a skeleton: every section and line present, each organ line showing its standard normal. The skeleton is the contract the placement engine works against. If the study is named more precisely later ("with and without contrast", "CT urogram"), the skeleton is swapped for the right variant and everything already placed is re-placed into it. ## One sentence, one placement Each transcribed sentence is sent to a placement model together with the current state of the report: every line, its label, whether it is still the default normal or already holds a dictated finding, and the last two sentences placed. The model returns a small set of operations: replace a line's default with the finding, append an additional finding to a line that already holds one, fill a bracketed blank, or file the sentence under *Additional findings*. Those operations are applied deterministically on your screen. Rules the engine follows, which you can rely on: - **A dictated finding always replaces the default normal** for its line. "Liver: unremarkable" becomes your finding, never both. - **A second finding for the same organ accumulates.** Dictating a second lung nodule adds a second line under Lungs; it never overwrites the first. - **Broad findings go on the broad line.** "Multilevel degenerative changes" lands on the spine line once, not fanned across every level. - **Named vessels go on their vessel line**, so a celiac stenosis dictated after an SMA finding joins it on the mesenteric line instead of being parked. - **Nothing is dropped.** A sentence the engine cannot place is kept verbatim under *Additional findings* with a review flag. ## Pauses, fragments and the transcriber's periods The live transcriber ends a segment whenever you pause and starts the next one with a capital letter. A breath in the middle of a finding therefore arrives as two "sentences": "7 mm right." and "Splenic lobe cyst." GigHz Precision AI does not trust those breaks. A segment whose tail cannot end a radiology sentence (a conjunction, a preposition, a bare laterality, a measurement with no noun) is held for a moment and joined to what follows; a segment that starts in lowercase or with a conjunction is treated as a continuation. The joined sentence is placed once, as one finding. A period you actually dictate is always honoured. ## It keeps working when the tab is in the background Radiologists dictate with PACS in front and the browser behind it. Browsers slow the timers of a hidden tab to once a minute, which would stall placement until the tab is brought forward. GigHz Precision AI runs its placement timers in a background worker that browsers do not throttle, so placement continues at full speed while you look at the images. ## Editing the transcript: rebuild, not patch If you edit the transcript (not just add to it), the report on screen still reflects the original wording and the old placements cannot be unpicked line by line. The application detects the divergence and offers **Rebuild report from edited transcript**: the skeleton is reset, every line of the edited transcript is placed again live, and the progress counter runs as it did the first time. Appending new sentences never needs a rebuild; they are placed incrementally. ## Finalizing **Finalize** waits for any sentence still in flight, composes the impression from the placed findings (see [Impression policy](/docs/impression-policy/)), and locks the report. A completeness check runs at this moment: every abnormal finding you dictated must appear somewhere in the report, and anything missing is added back. **Unfinalize** reopens the live report so you can keep dictating into it. ## What happens if the placement service is unavailable If the placement model cannot be reached, dictation is still captured and every sentence is parked under *Additional findings* with a visible warning. In that state the application deliberately does not compose an impression from unplaced text: the impression shows one honest line pointing at the parked findings, so a service outage can never produce a confident-looking but wrong impression. --- # Templates URL: https://precision.gighz.com/docs/templates/ Last reviewed: 2026-10-13 GigHz Precision AI ships with structured templates covering the common examinations in CT, MRI, ultrasound, radiography, fluoroscopy, nuclear medicine, breast imaging and interventional procedures, plus dedicated templates for LI-RADS liver imaging and multiphase protocols. The library currently carries 107 studies mapped to 101 live-fill bases, and grows with each release. ## How a study is routed The transcript is matched against an ordered table of study phrases. The table is deliberately specific-first: "CT urogram" is matched before "CT abdomen", "CTA chest for pulmonary embolism" before "CT chest". Contrast is detected separately ("with and without", "without contrast", "non-contrast") and selects the variant of the template whose technique line and organ defaults match. A typo in the study phrase does not cost you the template: common one-letter misrecognitions are normalised before matching. A matched study shows as a chip above the transcript. If nothing matches and the dictation is substantial, an AI router proposes an existing template; it can only choose from the real library, never invent one. ## Two layers: full templates and live-fill bases Each study has a full, detailed template (the classic "Create Report" output) and, for most, a simpler live-fill base: one line per organ or structure, written for the placement engine, with paragraph groupings so the finalized report reads as prose where that is conventional. Live dictation uses the base; typed generation uses the full template. Both share the same defaults and the same impression configuration, so a report reads the same whichever way it was produced. ## The lines A template line has a label, a default normal, and optional bracketed blanks. Lines marked *only if dictated* (limitations, lines and devices, comparison) stay out of the report unless you say something for them. A placement either replaces the default, appends to a dictated line, or fills a blank. Lines you edit by hand are protected: the engine never overwrites a hand-edited line. ## Making a template yours Above the transcript, after a study is detected, you see **Keep ACR** and **Make it mine**. Keep ACR leaves the standard template alone. Make it mine opens the structured editor for that study, where you can: - rename, reorder, add or remove sections and lines; - change a default normal to your own wording; - group lines into prose paragraphs or split them out; - switch the impression behaviour for that study (differentials when indeterminate, a standard follow-up recommendation, or none); - save the result as your default for that exact study and contrast variant. Your version is private to your account and is used automatically whenever that study is detected. The structured editor never loses text: every node keeps its original wording, and the editor can always fall back to plain text. ## Trigger phrases and personal templates A personal template can carry trigger phrases. When your dictation contains one, your template wins over the automatic detection, which is how a group's house style or an unusual protocol gets routed without renaming the study. ## Composable templates A few templates are composed from parts rather than fixed: universal CT and MRI bases that include only the regions you dictate, venograms by region, and multiphase liver, renal mass and adrenal washout protocols by phase. Those with a live-fill base (multiphase liver CT and MRI, adrenal washout) fill live; the purely region-composed ones are generated when you press Create Report. ## Interventional and procedure templates Procedure notes (biopsies, drains, lines, fistulograms, paracentesis and others) have their own templates and a universal IR base. "Load template" by voice, or the chip, brings one up, and the same live fill applies. --- # Voice commands URL: https://precision.gighz.com/docs/voice-commands/ Last reviewed: 2026-10-13 Commands are recognised only when the whole utterance is the command, so a finding that happens to contain the same words is never intercepted. The microphone keeps listening while the browser tab is in the background. ## Punctuation and structure | Say | Effect | |---|---| | "period", "comma", "colon" | The punctuation mark. "Period of" and "period in" are left alone, so "a period of six months" is not cut. | | "next line", "new line" | A line break in the transcript. | | "next paragraph" | A paragraph break. | ## Editing | Say | Effect | |---|---| | "scratch that", "delete that" | Removes the most recent utterance. During live fill, if that sentence has not been placed yet it is removed from the queue; if it has, its placement is undone. | | "revert liver to default", "reset the bones section" | Restores that line or section to the template's standard normal. | | "clear all", "start over" | Empties the transcript. | ## Templates | Say | Effect | |---|---| | "load template fistulogram" | Loads a built-in template by name (fuzzy matched). | | "load my template chest house style" | Loads one of your personal templates. | | "load SIR template paracentesis" | Loads a template from the Society of Interventional Radiology library. | | Naming a scoring system ("LI-RADS category", "Bosniak") | Runs that scoring macro on the dictation so far. | ## Report control | Say | Effect | |---|---| | "generate report", "create report" | During live fill: finalizes the live report. Otherwise: generates the report from the transcript. | ## Keyboard F4 toggles the microphone. Clicking any report line opens it for hand editing; Escape closes it. ## Tips - Name the study before anything else; that is what brings the template up. - You do not need to say "findings" or section names; the engine files each sentence by its content. - Say "period" when a sentence ends if you want to be explicit. If you forget, the engine decides from the words, not the pause. --- # Output options URL: https://precision.gighz.com/docs/output-options/ Last reviewed: 2026-10-13 Everything on this page is optional and off by default. Settings are remembered per account. ## Patient information Adds a short paragraph, after the impression, that explains the result in plain language and gives practical advice. It is written from the impression, uses no jargon, and is separated from the clinical report by its own heading so it can be left out when you paste. Intended for patient-facing portals where reports are released directly to patients. ## Billing suggestions Adds suggested CPT and ICD-10 codes derived from the study and findings. These are suggestions for your coding workflow and are not a billing determination. ## Signature Appends your signature block (name, credentials, any attestation text) to every report. Set once on the account page. ## Critical note The **Critical Note** button inserts a standard critical-findings communication paragraph (who was notified, when, by what means) for you to complete. It is a text insert; the application does not send the communication itself. ## Report history Your recent reports are available under **History** and can be reopened and edited. See [Data handling and retention](/docs/data-handling/) for what is stored and for how long. ## Feedback on a dictation **Report an issue with this dictation** at the bottom of a report sends a structured report to the developers: the study, which features were on, counts and timings, and your description of what went wrong. The text of the report itself and the indication are deliberately excluded from that payload; the application is built so that report content cannot ride along with feedback. If a case needs to be discussed in detail, email [signal@gighz.com](mailto:signal@gighz.com) and we will arrange a secure channel. ## Quality indicator After generation, a quality indicator summarises completeness: whether an indication was given, whether the impression addresses it, and whether optional sections were requested and produced. It is a prompt to review, not a score of clinical correctness. --- # Reference tools URL: https://precision.gighz.com/docs/reference-tools/ Last reviewed: 2026-10-13 Under the **More** menu beside the transcript are two reference tools that open in a panel without leaving the report. ## Protocol lookup A searchable library of imaging protocols: phases, contrast timing, coverage and reconstruction for CT, sequences and planes for MRI, and the standard protocol for common fluoroscopic and nuclear medicine studies. Ask in plain language ("how do I protocol a CT venogram of the pelvis", "MR enterography sequences") and the matching protocol is shown. The library is maintained with the application and updated with releases. ## Appropriateness lookup A search over the ACR Appropriateness Criteria: describe the clinical scenario and the panel shows which examinations are usually appropriate, may be appropriate, or are usually not appropriate, with the relative radiation level. Useful when a referrer asks which study to order, or when a requested study is not the right one for the question. ## SIR procedure templates The interventional template library follows the Society of Interventional Radiology's structured reporting elements for common procedures. Load one by voice ("load SIR template nephrostomy") or from the Templates panel. Procedure details, devices, specimens, complications and post-procedure instructions each have their own line, and the same live fill applies. ## Macros Macros are short phrases that expand into longer standard text. Define them under **Macros**; say or type the phrase and the expansion is inserted. Typical uses are a group's standard limitation sentence, a normal-chest paragraph, or a post-procedure instruction block. ## Dictation corrections Under **Commands** you can define word corrections that are applied to the transcript before placement: a drug name the transcriber mishears, a colleague's name, a local abbreviation. Corrections are per account. --- # Impression policy URL: https://precision.gighz.com/docs/impression-policy/ Last reviewed: 2026-10-13 The impression is the part of a report that is read, so GigHz Precision AI treats it differently from the findings. The findings are what you dictated, placed where they belong. The impression is composed, ordered and filtered by rules that are the same for every report and that you can read on this page. ## Your own impression always wins If you dictate an impression ("Impression: one, acute appendicitis…"), it is used verbatim. It is renumbered and image citations are removed; nothing else is touched, ordered or filtered. Everything below applies only when the impression is composed for you. ## Composition A composer model is given the dictated findings, the dictated indication and limitation if any, and the study type, with these instructions: 1. Determine the clinical question: the indication if dictated, otherwise inferred from the limitation or the study (a CT pulmonary angiogram is answering pulmonary embolism). 2. Answer that question first, explicitly, even when the answer is negative. 3. Then list the remaining findings by clinical severity. 4. A time-critical finding always becomes item one, above the indication. 5. Vascular studies must characterise the vessels. 6. Measurements and laterality are copied exactly; image and series numbers never appear. 7. Normals and pertinent negatives are excluded; every distinct abnormal finding is included; similar lesions are summarised as a group with a size range. ## Deterministic ordering The composer's lines are then ranked by a priority table that does not depend on the model. Active bleeding, solid organ injury, aortic catastrophe, tension physiology and intracranial haemorrhage sit at the top; torsion, bowel ischaemia and perforation, cord compression, acute stroke and limb ischaemia next; then abscess, mass, fracture, aneurysm, inflammatory processes, obstruction, thrombosis, effusions, nodules requiring follow-up; incidental and degenerative findings last. A term inside a negated clause ("no active extravasation") never ranks a line. "Chronic", "stable", "unchanged" and similar words demote a line out of the emergency tiers, so a chronic pulmonary embolism is not treated as a massive one. Simple cysts and small lymph nodes are gathered into one prose sentence rather than numbered lines; a cyst with features (septated, nodular, enhancing, complex) keeps its own line. ## What is kept out The rule, from the radiologists who shaped it: if an indication is given, the impression should answer it, and otherwise mention only what is urgent or needs follow-up. So the following never appear in a composed impression unless the same line carries an escalating term (complex, numerous, fractured, loosened, obstructing, new, increased, suspicious…) or the finding is itself the answer to the indication: - simple atelectasis; - unchanged degenerative or postoperative change of the bones, including intact hardware; - simple cysts (hepatic, renal, punctate); - atherosclerotic and other vascular calcifications; - diverticulosis without diverticulitis; - a collapsed or under-distended bladder; - artifact and limitation statements. A second tier, including uncomplicated hernias, lipomas, gallstones without cholecystitis, hepatic steatosis and reactive nodes, stays in the impression only when no indication was given or nothing else answers it. If every line is removed, the impression reads "No acute abnormality." ## Completeness check After composition, a backstop compares the impression against the dictated findings. Any abnormal finding missing from the impression is appended, through the same sanitiser and the same suppression rules, so the backstop can neither drop a finding nor reintroduce an incidental one. ## Per-template control Through **Make it mine** a template can set its own impression behaviour: differentials when a finding is indeterminate, a standard follow-up recommendation, or neither; and ordering by priority (default), as composed, or emergencies-only. ## Why this is deterministic on purpose The ordering and suppression tables are configuration files that ship with the application and are checked by the automated test suite on every release. The behaviour of the impression should not depend on a model's mood: a splenic laceration with active extravasation is item one every single time. --- # Clinical decision support URL: https://precision.gighz.com/docs/clinical-decision-support/ Last reviewed: 2026-10-20 GigHz Precision AI embeds the common radiology classification systems so that the category and its standard recommendation are derived from what you dictate, in your words, while you dictate. The systems are applied as structured rules, not free-form model opinion, and every one of them is confirmed by you before it enters the report. ## Systems supported | System | Applies to | What is derived | |---|---|---| | LI-RADS v2018 | Liver observations in patients at risk for hepatocellular carcinoma, CT and MRI | Category (LR-1 to LR-5, LR-M, LR-TIV) from size, arterial phase hyperenhancement, washout, enhancing capsule and threshold growth; ancillary features | | Bosniak 2019 | Cystic renal masses, CT and MRI | Class I to IV from wall and septa thickness, enhancement, nodularity and calcification | | Fleischner 2017 | Incidental pulmonary nodules | Follow-up recommendation from size, solidity, number and risk | | Lung-RADS | Lung cancer screening CT | Category and management | | TI-RADS (ACR 2017) | Thyroid nodules on ultrasound | Points per feature (composition, echogenicity, shape, margin, echogenic foci), the TR level, and the FNA or follow-up size threshold; unstated features are reported as not scored | | PI-RADS v2.1 | Prostate MRI | Assessment category by zone | | O-RADS | Adnexal lesions on ultrasound and MRI | Risk category | | PECARN | Paediatric head trauma | Decision rule support | ## Which run live, and which run at generation LI-RADS, Bosniak, Fleischner and TI-RADS run **live**: a panel appears beside the report while you dictate, showing the parsed inputs and the derived category. PI-RADS, Lung-RADS, O-RADS and PECARN are applied when the report is generated and through the scoring voice macros (say the system's name after describing the finding). The same rule set is used in every path. ## How a system is triggered Each system has a detector that watches the dictation for its context. LI-RADS, for example, requires both a liver observation and an at-risk liver (cirrhosis, chronic hepatitis B, or a stated indication of hepatocellular carcinoma surveillance); a liver lesion in a healthy liver never triggers it. Bosniak needs a cystic renal lesion with the features described. The detectors run on the live transcript, and a non-blocking panel appears beside the report with the parsed inputs and the derived category. ## Confirming, correcting and dismissing The panel shows exactly which features were parsed from your words ("arterial phase hyperenhancement: yes; washout: not stated"). You can correct an input, confirm the category, or dismiss the panel. A confirmed category is added to the impression as a standard assessment line with the system's recommendation; a dismissed one is not mentioned anywhere. The panel never interrupts dictation and never opens a modal dialog while the microphone is live. ## One rule set per system Each system has exactly one implementation inside the application: one feature extractor and one decision grid, used by the live panel, the voice macro ("LI-RADS category") and the generated-report path alike. The test suite enforces that a feature regex can never match its own negation ("non-enhancing" must not read as "enhancing"), that the grids agree with the published tables on a corpus of worked cases, and that the live panel and the generated report derive the same category from the same prose. ## Canonical wording When a category is confirmed, the report receives the system's canonical statement, assembled from configuration files that ship with the application, so the wording is consistent across users and releases. Your own dictated wording for the observation itself is preserved. ## Watch them run Three of the live systems, recorded on the application with synthetic dictations. Each video is click-to-play; see [Demo videos](/docs/demo-videos/) for all of them. @video D7MZogmzjEs @video uaPC9XEtAEQ @video 2oK3T3asdH4 ## What the systems are not They are decision support, not decisions. The category is derived from the features you dictated; if a feature was not stated it is not assumed. Review the parsed inputs before confirming, exactly as you would review a trainee's categorisation. --- # HIPAA, Business Associate Agreements and PHI URL: https://precision.gighz.com/docs/hipaa-and-baa/ Last reviewed: 2026-10-13 Radiology groups evaluating AI reporting tools ask the same three questions: where does the data go, who has signed what, and does the product need patient identifiers at all. This page answers them plainly. It describes the architecture and agreements as they are; it is not legal advice, and your compliance officer should review it against your own policies. ## The short version - GigHz Precision AI runs in a **HIPAA-eligible AWS environment under a signed Business Associate Agreement** between GigHz LLC and Amazon Web Services, with encryption in transit and at rest. - Dictation audio and report text are processed through **OpenAI's API services under agreements covering that processing**; content sent to the API is not used to train OpenAI's models under OpenAI's API terms. - The product **does not require patient identifiers**, and the Terms of Service prohibit entering them. A report is drafted from the findings, the indication and the study type, none of which needs a name, record number or date of birth. - There is **no advertising and no third-party analytics** on the application, and the feedback system is built so report text cannot leave with it. ## Why "does not need PHI" is the important design choice Most of the HIPAA surface of a reporting tool comes from identifiers travelling with the clinical text. GigHz Precision AI was designed so that the clinical content is useful without them: the template is chosen from the study name, the findings are placed from their anatomy, and the impression is composed from the findings and the indication. Nothing in the workflow asks for who the patient is. The application instructs users not to enter identifiers, the Terms make that a condition of use, and the privacy notice explains it. A user who follows the Terms is sending de-identified clinical text. That is also why the vendor agreements are described carefully: a Business Associate Agreement between GigHz LLC and its vendors governs GigHz LLC's obligations to those vendors; it does not by itself authorise an individual user to submit PHI. For a covered entity that wants to use the product with identifiable data, the right instrument is a Business Associate Agreement between that entity and GigHz LLC. Contact [signal@gighz.com](mailto:signal@gighz.com) to discuss one. ## The data flow, step by step | Step | What moves | Where | Protection | |---|---|---|---| | Dictation | Audio from your microphone | Your browser to the transcription service | TLS; audio is streamed, not stored by GigHz Precision AI | | Placement | Each sentence and the current report state | Application server to the language model API | TLS; API terms exclude training; no identifiers needed | | Impression | The placed findings and indication | Same | Same | | Report history | The finished report text | Application server, encrypted storage, per-account | Encrypted at rest; retrievable only by that account | | Feedback | Metadata (study, feature flags, counts, your comment) | Application server | Report text and indication are structurally excluded | | Email | Account emails (verification, password, onboarding) | Amazon SES from the application's own domain | DKIM and SPF; no report content in any email | ## Infrastructure and agreements - **Hosting:** Amazon Web Services, United States region, HIPAA-eligible services only, under the AWS Business Associate Addendum. Compute, storage and email are all inside that account. - **Language and transcription models:** OpenAI API services under agreements covering that processing. Content is not used to train models under the API terms. - **Edge:** the application domain is served directly from AWS over TLS; it is deliberately not proxied through a content delivery network, so no third party terminates TLS for clinical traffic. See [Security architecture](/docs/security-architecture/). - **No other processors** receive clinical content. Payment processing (Stripe) receives billing details only; the Google Sheet that records signups receives account metadata only. ## What a covered entity should check 1. **Identifier policy.** Confirm your radiologists understand the product does not need identifiers and that your workflow does not paste them in. 2. **Agreements.** If you intend to use identifiable data, request a BAA with GigHz LLC. 3. **Access control.** Each radiologist has an individual account; there are no shared logins, and administrative functions are separated from clinical use. See [Security architecture](/docs/security-architecture/). 4. **Retention.** Review [Data handling and retention](/docs/data-handling/) against your own retention schedule; account deletion is available on request. 5. **Audit.** Access to the application is logged by account; email delivery is logged; feedback is logged without content. ## What we do not claim We do not describe the product as "HIPAA certified" because no such certification exists, and we do not describe it as "HIPAA compliant" as a blanket statement, because compliance is a property of how a covered entity and its business associates operate together, not a label a vendor can apply to itself. What we can state is the architecture, the agreements and the design choices above, and we keep this page current as they change. ## Related - [Security architecture](/docs/security-architecture/) - [Data handling and retention](/docs/data-handling/) - [AI models and limitations](/docs/ai-models-and-limitations/) - Terms of Service (version PRECISION-TOS-2026-08-31-v0.9) and Privacy Notice (version PRECISION-PRIV-2026-08-31-v0.9), shown and accepted at signup --- # Security architecture URL: https://precision.gighz.com/docs/security-architecture/ Last reviewed: 2026-10-13 This page describes the production environment as it is built. It is written for the person at a radiology group or hospital who has to sign off on a new tool. ## The boundary The application runs on a single-tenant environment inside GigHz LLC's AWS account, in a United States region, using only HIPAA-eligible services. The web application, its database, the email service and the deployment pipeline all live inside that account. There is no shared hosting and no third-party platform between the browser and the application. ## Transport The application domain resolves directly to the AWS environment and TLS terminates there. It is deliberately not proxied through a content delivery or DDoS-mitigation network: proxying would mean a third party decrypts clinical traffic in transit. The company's marketing site is a separate system on a separate domain and is proxied in the ordinary way; nothing clinical passes through it. ## No third-party code on the clinical page Every script the application page loads is served from the application itself. The JavaScript libraries it depends on are vendored and pinned to exact versions; nothing is loaded from a public CDN. There is no advertising, no third-party analytics and no tag manager. The automated test suite fails the build if a public CDN host or a floating dependency version appears in the plugin. ## Accounts and roles - Each radiologist has an individual account created through email verification and acceptance of the Terms. There are no shared or demo logins on the production site. - Clinical users cannot reach the administrative console; administrative URLs are locked to administrators, and the administrative toolbar is not shown to clinical users. - Administrative functions (settings, rollout of features, review queues) are separated from the clinical application, and anything intended for administrators is not rendered to clinical users. - Author and user listings are disabled, so account names cannot be enumerated from the site. - Login hardening: generic failure messages, no username disclosure, rate limiting at the web server. ## Secrets API credentials are stored server-side and never shipped to the browser; the browser calls the application, which calls the model services. Keys are entered by the owner directly into the administrative console and are never transmitted through chat, email or support channels. Email is sent by signing requests with the server's own IAM role, so there is no email password anywhere in the system. ## Email Account email (verification, password links, onboarding, billing notices) is sent through Amazon SES from the application's own domain with DKIM signing, SPF and DMARC. Bounces and complaints are delivered to the operations mailbox. No email ever contains report content. Every message is logged (recipient, subject, status, time) for delivery auditing. ## Backups and recovery The environment is backed up on a schedule within the same AWS account; backups inherit the account's encryption and access controls. Recovery is from backup to the same environment. ## Change control and testing Every release passes a deterministic test suite before it can be packaged (currently several hundred invariants covering routing, placement, impression rules, clinical scoring, the PHI boundary of the feedback payload, the CDN gate and the deployment health of the live site), plus mutation testing that plants defects to prove the checks can fail. A nightly run re-tests the deployed site and files any regression. See [Quality assurance](/docs/quality-assurance/). ## Reporting a security concern Email [signal@gighz.com](mailto:signal@gighz.com) with "security" in the subject. We acknowledge within two business days. --- # Data handling and retention URL: https://precision.gighz.com/docs/data-handling/ Last reviewed: 2026-10-13 This page is the inventory. The [HIPAA, BAAs and PHI](/docs/hipaa-and-baa/) page covers the agreements and the identifier policy; the Privacy Notice (version PRECISION-PRIV-2026-08-31-v0.9) is the governing document. ## What is stored | Data | Purpose | Where | Retention | |---|---|---|---| | Account: name, email, role, specialty, institution, password hash | Running your account | Application database | While the account exists | | Terms and Privacy acceptance records (version, time, typed name) | Evidence of acceptance | Application database | Retained after account deletion, as the notice states | | Trial and subscription state | Billing | Application database and Stripe | While the account exists; Stripe per its own policy | | Report history | Your recent reports under History | Encrypted per-account storage on the application server | Until you delete them or the account is deleted | | Personal templates, macros, corrections, preferences | Your settings | Application database | While the account exists | | Feedback on a dictation | Improving the product | Application database and an operations spreadsheet | Metadata only; see below | | Email log (recipient, subject, status, time) | Delivery auditing | Application database | Operational period, then purged | | Usage counters per account | Abuse detection and capacity | Application database | Aggregated; never block a user | ## What is processed but not stored - **Dictation audio** is streamed to the transcription service and is not recorded by GigHz Precision AI. - **Each sentence and the report state** are sent to the placement and composition models for the duration of the request. The application keeps the resulting report in your history; it does not keep the prompts. - **Content sent to OpenAI's API** is not used to train models under the API terms. ## What is never collected - Patient identifiers: the product does not ask for them and the Terms prohibit entering them. - The text of your report inside feedback: the feedback payload carries a character count, never an excerpt, and the test suite enforces that. - Advertising or third-party analytics identifiers: there are none on the application. ## Where it lives Everything above except Stripe's billing records and the operations spreadsheet lives inside GigHz LLC's AWS account, on encrypted storage, in the United States. Stripe receives payment details and the account email; the spreadsheet receives signup and feedback metadata, never clinical text. ## Deleting your data - Individual reports can be deleted from History at any time. - To delete your account, email [signal@gighz.com](mailto:signal@gighz.com) from the account's address. We respond within 30 days, as the Privacy Notice states. Account records and report history are removed; acceptance records are retained. - Cancelling a subscription does not delete the account; the account remains readable until you ask for deletion. ## Access to your data by GigHz LLC Operational access is limited to the owner and is used for support and incident response. We do not read reports in the course of normal operation, and we do not use report content to train models. If a support case requires looking at a specific report, we ask you first. --- # AI models and limitations URL: https://precision.gighz.com/docs/ai-models-and-limitations/ Last reviewed: 2026-10-13 GigHz Precision AI uses large language models for three narrow jobs and deterministic rules for everything that must not vary. Knowing where the line is tells you where to look when you review a report. ## Where models are used | Job | Model role | Deterministic layer around it | |---|---|---| | Transcription | Speech to text, streamed | Silence-filler gate (a lone "okay" during a pause is discarded), dictation corrections, spoken punctuation | | Placement | For each sentence, choose the template line and the operation | Fixed set of allowed operations; a dictated finding always replaces a default; hand-edited lines are protected; unplaceable text is kept verbatim | | Impression composition | Write numbered lines from the findings and indication | Priority ordering, incidental suppression, image-citation removal, de-duplication, completeness backstop | | Study routing fallback | Propose a template when phrase matching fails | May only choose from the real library | | Reference lookups | Answer protocol and appropriateness questions | Scoped to the curated libraries | Clinical scoring systems (LI-RADS, Bosniak, Fleischner and the rest) are not model output: their features are parsed by rules and their categories come from the published tables, implemented once each. The model is only ever asked to place the sentence; it is never asked to categorise. ## Model selection The application uses current OpenAI models through the API, with a stronger model automatically selected for complex or high-risk cases (an active scoring-system case, or dictation that flags as complex). The specific models are configured by the administrator and change as better ones become available; the documentation does not pin them because the behaviour that matters is governed by the deterministic layers above, which are tested on every release. ## Known failure modes, and what catches them - **A sentence lands on the wrong line or is parked.** Visible on screen as you dictate; parked text is flagged. A named-vessel finding that the model parks is redirected to its vessel line by rule. - **A sentence is split by a pause.** Fragments are re-joined before placement; the engine decides sentence boundaries from the words. - **The composer omits a finding.** The completeness backstop re-adds any dictated abnormal finding missing from the impression. - **The composer includes an incidental finding.** The suppression rules remove it, at every door, after the backstop. - **The transcriber invents a filler during silence.** Filler-only utterances are discarded before they reach the transcript. - **The placement service is down.** Findings are parked with a warning and no impression is composed from unplaced text. - **A wrong number.** Measurements are copied, never computed; the test corpus checks that dictated numbers appear verbatim. Review every measurement anyway. ## What the model cannot do It does not see the images. Every finding in a report was dictated by you; the software organises, it does not observe. It does not know the patient's history beyond the indication you state. It does not check your findings against each other for clinical consistency beyond the scoring systems. ## Regulatory position GigHz Precision AI, as sold, is a dictation-based drafting and decision-support tool: it organises text the radiologist dictates and applies published classification rules to that text. It does not analyse medical images. Image-analysis features that generate findings from pixels are a regulated device category in the United States and are not part of the product; any such capability is confined to internal research by the developer and is not available to users. ## The reviewer's responsibilities The Terms of Service (version PRECISION-TOS-2026-08-31-v0.9) require that every report be reviewed by the radiologist before use. In practice that means: read every line of the findings against what you intended; check every measurement and laterality; read the impression as the referrer will; confirm or dismiss any scoring panel; and treat *Additional findings* as a to-do list. The application is built to make that review fast. It is not built to make it optional. --- # Quality assurance URL: https://precision.gighz.com/docs/quality-assurance/ Last reviewed: 2026-10-13 A reporting tool that changes behaviour between releases is worse than one that never improves. GigHz Precision AI is developed under a test regime designed so that a fix, once made, cannot quietly come undone. ## The build gate A release cannot be packaged unless every deterministic check passes against the source tree. The checks are not unit tests of functions in isolation; they are invariants about the product's behaviour, extracted from the real code and run against the real configuration: - **Routing:** each study phrase in the corpus resolves to the template it should, contrast is bucketed correctly, no routing row points at a template that does not exist, and a one-letter misrecognition does not cost the template. - **Placement engine:** a dictated finding can never be silently destroyed; a second finding on the same line accumulates; a 5 mm nodule and a 5 cm mass are distinct; hand-edited lines are protected; named vessels are rescued from the parking section through every door. - **Impression:** emergencies lead; chronic findings are demoted; incidental findings are suppressed at every door after the backstop; the impression never repeats itself; dictated impressions are never rewritten. - **Clinical scoring:** one implementation per system; feature regexes cannot match their own negation; the grids agree with the published tables on worked cases; the live panel and the generated report derive the same category. - **Boundaries:** the feedback payload cannot carry report text; no script loads from a public CDN; administrative features are not rendered to clinical users; secrets never ship to the browser. - **Deployment health:** the live site answers, reports its version, has its model configured, serves its vendored libraries, and matches the packaged build. At the time of writing the suite runs several hundred such checks on every build and every night. ## Testing the tests A check that cannot fail is worse than no check, because it reads as coverage. For every family of check, a mutation test plants the defect the check claims to detect and confirms the check goes red. A check that stays green on its planted defect is reported as blind and the build fails. ## The nightly corpus replay A corpus of dictations, most of them harvested from real field failures, is replayed every week against the live site through the same placement and composition paths a radiologist uses, and the output is scored: did each finding land on its line, were measurements kept verbatim, does the impression mention what it must and exclude what it must not, is the numbering sound. On the other nights the free deterministic checks run against the live site, together with a replay of only the cases tied to open issues, so a regression is still seen the next morning. ## The issue registry Every failure is fingerprinted so that the same defect, seen on a different patient weeks later, collapses to one identity. The registry tracks each identity from open, to fixed, to verified after consecutive clean runs, and flags any identity that fails again after being fixed as a regression, which is the loudest thing in the report. Several distinct identities sharing a subsystem are flagged as a theme: a sign that the mechanism, not the symptom, needs changing. ## From field report to permanent check When a radiologist reports a problem, the dictation that caused it becomes a corpus case with explicit expectations, the fix is made, the check that would have caught it is added to the suite, and the mutation test proves the new check can fail. The report in this documentation that explained a behaviour was, in most cases, written after exactly that cycle. ## What is not tested automatically Clinical correctness of a composed impression on a novel case is judged by the reviewing radiologist, not by the suite; the suite guarantees the rules were applied, not that the rules are medically complete. That is why the Terms require review, and why feedback from working radiologists is the input the suite grows from. --- # Plans, trials and billing URL: https://precision.gighz.com/docs/pricing-and-trials/ Last reviewed: 2026-10-13 There are two prices and two trials, and one rule decides which trial you get. ## Prices | Plan | Price | Who | |---|---|---| | Attending | $99 per month | Practising radiologists | | Resident / Fellow | $49 per month | Radiologists in training, with a self-reported graduation year | Prices are monthly, billed by card through Stripe, and cancellable at any time from the account page. There is no annual plan and no coupon system; the trials below are the only promotional terms that exist. ## Trials | Email at signup | Trial | Card required | |---|---|---| | Institutional (hospital, university, practice domain) | 90 days | No | | Any other email | 30 days | Yes, on file, not charged during the trial | The institutional list covers academic medical centres, health systems and known practices; a domain that looks institutional but is not on the list is placed in a short review queue rather than refused, and most are approved the same day. Personal mailbox domains always take the card trial. ## What happens at the end of a trial - **Institutional trial:** you receive reminders at days 7, 30 and 60, and a notice seven days before expiry with a link to start a subscription. If no subscription is started, the account is locked after a 7-day grace period and can be reactivated at any time by subscribing. - **Card trial:** a notice seven days before expiry; the subscription starts automatically on the card on file unless you cancel first. Receipts go to the email you choose on the account page. ## The trainee rate after graduation A resident or fellow enters a graduation year at signup. In August of that year the trainee rate is automatically extended for one further year, and in August of the following year the subscription moves to the attending price. You are emailed at each step. If the automatic change fails for any reason, it is made by hand and you are told. ## Managing billing The account page links to the Stripe customer portal, where you can update the card, see invoices and cancel. Cancelling stops future charges at the end of the current period; it does not delete your account or your report history. ## Groups Group and departmental arrangements, including a Business Associate Agreement where required, are handled directly. Email [signal@gighz.com](mailto:signal@gighz.com). ## What we do not promise The Terms of Service are explicit that availability, response time, compatibility with any PACS or RIS, and the availability of any particular AI model are not guaranteed. The product is sold month to month precisely so that you are never locked in. --- # Your account URL: https://precision.gighz.com/docs/account/ Last reviewed: 2026-10-13 ## Creating an account 1. Go to [https://precision.gighz.com/signup/](https://precision.gighz.com/signup/) and enter your name, email, institution, role (attending, fellow, resident, other) and specialty. 2. Click the verification link in the email. If it does not arrive within a few minutes, check spam; the sender is the application's own domain. 3. Read the Terms of Service and the Privacy Notice. Both must be scrolled to the end before the acknowledgments unlock. Three separate acknowledgments and your typed name are required; these are recorded with the document versions (PRECISION-TOS-2026-08-31-v0.9, PRECISION-PRIV-2026-08-31-v0.9) as evidence of acceptance. 4. Your account is created, a separate email carries the password-setting link, and the onboarding email explains the first dictation. The email you sign up with decides your trial; see [Plans, trials and billing](/docs/pricing-and-trials/). ## Signing in Sign in at [https://precision.gighz.com/](https://precision.gighz.com/) with your email and password. The application is at [https://precision.gighz.com/app/](https://precision.gighz.com/app/). There is no shared login and no demo account on the production site. ## Password Set the password from the link in the welcome email. To change or reset it later, use **Reset password** on the account page; a link is emailed to your account address. Passwords are stored as salted hashes and are never emailed in clear text. ## Display name The name shown in the application greeting and used in your signature block. Change it on the account page. ## Notification email If you signed up with an institutional address that filters external mail, you can set a preferred notification address on the account page. A confirmation link is sent to the new address before it takes effect. Billing receipts and trial notices then go there; the login email does not change. ## The account page - Trial or subscription status and the date it changes - Plan chooser (attending or trainee) and the Stripe portal link - Password reset - Display name and notification email - Links to the Terms and Privacy Notice you accepted ## Deleting an account Email [signal@gighz.com](mailto:signal@gighz.com) from the account address. See [Data handling and retention](/docs/data-handling/) for what is removed and what is kept. --- # Frequently asked questions URL: https://precision.gighz.com/docs/faq/ Last reviewed: 2026-10-13 ## Does GigHz Precision AI need patient identifiers? No. The report is drafted from the study name, the findings and the indication. The product does not ask for a name, record number or date of birth, and the Terms prohibit entering them. See [HIPAA, BAAs and PHI](/docs/hipaa-and-baa/). ## Is it HIPAA compliant? We do not use that phrase, because HIPAA compliance describes how a covered entity and its business associates operate together, not a label a vendor can apply to itself. What we state is the architecture: a HIPAA-eligible AWS environment under a signed Business Associate Agreement, encryption in transit and at rest, no advertising or third-party analytics, and a product designed not to need PHI. Covered entities that want to use identifiable data can request a BAA with GigHz LLC. ## Where is my data stored? In GigHz LLC's AWS account in the United States, on encrypted storage. Report history is kept per account until you delete it. Dictation audio is streamed and not recorded. See [Data handling and retention](/docs/data-handling/). ## Does it integrate with my PACS or RIS? No, by design. It runs in a browser beside your PACS; you dictate, review, finalize and copy the report into your reporting system. There is nothing to install and no IT project. The Terms make no compatibility promises. ## Does the AI read the images? No. Every finding in a report was dictated by you. The software places and organises what you said and applies published classification rules; it does not observe the images. ## How accurate is it? Accuracy is the wrong frame for a drafting tool: the report contains your findings in your words, placed on template lines. What the software decides is where a sentence belongs, how the impression is ordered and what is kept out of it, and those decisions follow published rules that are tested on every release. Every report is a draft for your review, and the Terms require that review. ## Which AI models does it use? Current OpenAI models through the API, with a stronger model selected automatically for complex cases. The models are configurable and change over time; the deterministic rules around them are what is tested. See [AI models and limitations](/docs/ai-models-and-limitations/). ## What does it cost? $99 per month for attendings and $49 per month for residents and fellows, monthly, cancel any time. Institutional email addresses get a 90-day trial with no card; other addresses get 30 days with a card on file. See [Plans, trials and billing](/docs/pricing-and-trials/). ## Can I change the template wording? Yes. Click **Make it mine** above the transcript to edit any template's sections, lines and defaults; your version becomes your default for that study. See [Templates](/docs/templates/). ## Why does the impression leave things out? Because the radiologists who shaped the rules asked for an impression that answers the clinical question and mentions only what is urgent or needs follow-up. Simple atelectasis, unchanged degenerative change, simple cysts and vascular calcification are suppressed unless something about them is escalated. See [Impression policy](/docs/impression-policy/). ## What happens if the AI service is down? Dictation is still captured. Findings are parked under Additional findings with a visible warning, and no impression is composed from unplaced text. ## Can residents use it? Yes, at the trainee price, with a self-reported graduation year. The trainee rate continues for one year after graduation. The Terms require that reports be reviewed by a radiologist authorised to sign them. ## How do I report a problem? Use **Report an issue with this dictation** on any report, which sends metadata without the report text, or email [signal@gighz.com](mailto:signal@gighz.com). Field reports become corpus cases and permanent checks; see [Quality assurance](/docs/quality-assurance/). ## Who is behind GigHz Precision AI? GigHz LLC, a California company founded by a practising radiologist, and a member of [NVIDIA Inception](https://www.nvidia.com/en-us/startups/). Contact [signal@gighz.com](mailto:signal@gighz.com). ## How do I delete my account? Email [signal@gighz.com](mailto:signal@gighz.com) from the account address; we respond within 30 days. Acceptance records are retained; everything else is removed. --- # Glossary URL: https://precision.gighz.com/docs/glossary/ Last reviewed: 2026-10-13 **Additional findings.** The catch-all section at the end of the findings where any dictated sentence that the placement engine could not file with confidence is kept verbatim, with a flag for review. Nothing dictated is ever discarded. **Backstop (completeness).** The check at finalization that compares the impression against the dictated findings and appends any abnormal finding that is missing, through the same filters as the composer. **Composable template.** A template assembled from parts according to the dictation: regions for universal CT and MRI bases, phases for multiphase liver, renal mass and adrenal protocols. **Default normal.** The sentence a template line shows when nothing has been dictated for it ("Liver: Normal in size and attenuation. No focal lesion."). A dictated finding replaces it. **Finalize.** Locking the live report: in-flight sentences are placed, the impression is composed, the completeness backstop runs, and the report is ready to copy. Unfinalize reopens it. **Impression suppression.** The rule set that removes incidental findings (simple atelectasis, unchanged degenerative change, simple cysts, vascular calcification and others) from a composed impression unless escalated or unless they answer the indication. **Indication.** The clinical question for the study, dictated or inferred. The composed impression answers it first. **Live-fill base.** The simplified, one-line-per-organ version of a template used by live placement, with paragraph groupings for the finalized report. **Live report fill.** Placing each dictated sentence on its template line as it is spoken, so the report assembles during dictation. **Make it mine.** The control that opens the structured template editor for the detected study and saves your version as your default. **Placement.** The operation of filing one dictated sentence onto the right template line: replace the default, append to a dictated line, fill a blank, or park under Additional findings. **Priority ordering.** The deterministic ranking applied to composed impression lines: emergencies first, incidental findings last, chronic findings demoted. **Rebuild.** Re-placing every line of an edited transcript into a fresh skeleton of the same template, live, with the progress counter. **Scratch that.** The voice command that removes the most recent utterance and undoes its placement. **Seam.** The join between two transcriber segments that were one spoken sentence. The engine re-joins a segment whose tail cannot end a sentence with the segment that follows. **Skeleton.** The empty template shown as soon as the study is named: every section and line, each with its default normal, before any finding has been placed. **Structured template editor.** The editor behind Make it mine: a parsed model of the template (sections, lines, defaults, blanks) that always serialises back to the exact original text when unchanged. **Trainee rate.** The resident and fellow price, continued for one year after the stated graduation year and then moved to the attending price automatically. **Trial lane.** The rule that assigns a signup to the institutional trial (institutional email, no card) or the card trial (any other email, card on file). --- # Release notes URL: https://precision.gighz.com/docs/release-notes/ Last reviewed: 2026-10-13 The version running now is **13.35.3**. Internal fixes and test-suite changes are not listed; behaviour a radiologist would notice is. ## 13.3x series (September and October 2026) - Live placement keeps working at full speed while the browser tab is behind PACS. - Fragments split by a pause ("7 mm right." + "hepatic lobe cyst.") are re-joined before placement; the transcriber's own periods are no longer trusted as sentence ends. - Editing the transcript offers a live rebuild of the whole report instead of re-placing edited lines on top of the old ones. - Impression policy: incidental findings (simple atelectasis, unchanged degenerative change, hardware, simple cysts, vascular calcification, diverticulosis, artifact statements) are kept out of composed impressions unless escalated or unless they answer the indication. - Near-duplicate sentences (a tense or typo variant of a line already placed) no longer appear twice. - Multiphase liver and adrenal washout protocols now fill live. - A named vessel finding that would have been parked is redirected to its vessel line through every path. - The blank line between the impression and the patient-information paragraph is preserved. - Nightly quality assurance moved to the production environment; paid corpus replay weekly, deterministic checks nightly. ## 13.2x series (August and September 2026) - Production moved to a single-tenant AWS environment with direct TLS, vendored libraries and no third-party scripts. - Self-serve signup with email verification, the Terms ceremony, institutional and card trials, attending and trainee prices, and automatic trainee-rate handling after graduation. - Onboarding email and the illustrated Getting started guide. - Silence fillers invented by the transcriber during pauses are discarded. - Incidental benign findings no longer lead an impression above a real finding; grouping and ranking apply to generated reports as well as live ones. - A placement-service outage parks findings with a warning and never composes an impression from unplaced text. - A public health probe lets the nightly tests confirm the live site is configured. ## 13.1x series (summer 2026) - Live report fill with the simple live-fill bases, the live impression draft, the switch-template-mid-fill control, and the structured template editor behind Make it mine. - One implementation per clinical scoring system, shared by the live panel, voice macros and generated reports; Bosniak 2019 and LI-RADS v2018 grids verified against worked cases. - Feedback payload made metadata-only. - Fragment coalescing, dangling-tail joins and multi-finding lines.