Second Opinion: AI Review of Case Material
How a physician can obtain an AI second opinion on reports, laboratory results, images, and the clinical case in the /diagnostics section — steps, conditions, limits, and boundaries.
What it is and who can access it
"Second Opinion" is a bedside tool for the physician. You open a "case" for a specific patient situation, add materials (report text, a photograph of a form, a PDF, an image, laboratory values), and the AI reviews them: what to re-verify, what must not be missed, what is missing from the data. The conclusion for the case is written and signed by the physician.
- Where: the physician's side menu → "Decision Support" group → "Second Opinion". The address is
/diagnostics; an individual case page is/diagnostics/cases/<id>. - Who: only users with the physician role. The section is deliberately closed to patients: a review without a physician would turn into self-diagnosis.
- Residents and students may only manage cases that are not linked to a clinic patient (external or anonymous cases).
- Plan: on the free physician plan (Free) the AI is disabled — you can create a case, but recognition and review will not start, and the system will suggest purchasing a subscription. More on limits below and in
/docs/plans.
The main rule: the decision belongs to the physician
- A case is closed by your conclusion, not by the review. Without your own conclusion (at least a few words) the case cannot be closed.
- AI text is never inserted into your conclusion automatically.
- Every AI finding is marked as supportive: "This is a supportive review, not a diagnosis. The physician makes and signs the decision." The notice is immutable and accompanies the data into any export.
- The assistant does not make diagnoses and does not prescribe treatment. A system that asserts a diagnosis about a living patient is a medical device subject to separate regulation; this module deliberately remains supportive.
Which investigations can be reviewed
13 categories:
- the clinical case as a whole (complaints, history, course — without a separate investigation);
- laboratory tests;
- radiography, CT, MRI, ultrasound;
- ECG, echocardiography, EEG, auditory brainstem response (ABR/BERA);
- endoscopy;
- histology and cytology;
- dermatology — from the text of the description (skin examination, dermoscopy); the model does not read skin photographs.
Each category has its own verification protocol: a list of "Checked" items and "Must not be missed" items (red flags). The protocol can be opened on the case page with the "What is checked" button even before submission — a hidden protocol would mean trusting blindly.
Three types of review operate internally:
- by investigation report — the text of a CT, MRI, ultrasound, echocardiography, radiography, ECG, EEG, ABR (BERA), endoscopy, histology, or dermatology report is reviewed according to the protocol for that modality;
- by laboratory values — the numbers are compared by the program according to rules, while the model explains their meaning; this is why there are no errors in the figures;
- by clinical case — the differential list and what further workup is needed.
Which categories are launched is determined by the composition of the materials in the case.
Step-by-step workflow
Step 1. Create a case
- Open
/diagnostics. - Enter a case title (for your use only, so you can find it in the list; for example, "Chest CT, focal lesion").
- Optionally, a patient label: "Patient K." or a chart number. The name is not needed — the review does not depend on it.
- Click "Create" — the case page will open.
- Write your question ("what exactly concerns you?") and the clinical picture: age, sex, complaints, duration, course, what has already been done and ruled out. Do not enter full names, dates of birth, or document numbers. Fields are saved when you leave them.
The more precise the question, the more useful the answer: the AI responds primarily to it.
Step 2. Add materials
Each material is added separately, in three actions:
- Select the investigation type (CT, MRI, laboratory tests…). This determines the protocol. If it is a case description without an investigation, leave "Clinical case".
- Enter the material in one of the following ways:
- Paste text — the radiologist's report, a discharge summary, a description;
- Photograph of a form or a PDF — the AI will recognize the printed text and insert it into the field;
- An image (JPG, PNG, or a DICOM file) — the "Choose an image" button: the AI will look at the image itself and describe what is visible on it;
- Laboratory tests by values — "Enter values": name, value, units, and reference range from the form (copy the reference range from the form — it differs between laboratories). You can add your own value that is not on the list.
- Review the text and click "Add to case". Until you click, the recognized text will not enter the review.
What has been added is visible in the "Already in the case" list. Unnecessary material can be removed as long as the case is not closed.
Step 3. Start the review
Click "Review". Before the first run for a case, the system will ask you to confirm two conditions (see below) — once per case.
The review runs as tasks, one per category. Each task shows its status: queued, in review, ready, failed, skipped. This usually takes a few minutes; you can close the page — the result will be saved in the case. If some tasks fail, a line "Could not review: …" lists those categories, with a "Try again" button — it restarts all failed tasks at once; the successful ones are not touched. If a review "hangs" for more than 15 minutes, after the page is refreshed such a task is marked as failed and can be run again.
Step 4. Read the result and write your conclusion
- At the top is the "Case summary": the answer to your question.
- "What is missing from the description" — not conclusions about the patient, but what is worth adding so that the review becomes more complete.
- Findings — each with: importance (critical / important / note), confidence (high / medium / low — in words, not percentages), rationale, what to do, and the protocol item. Critical findings always come first in the list.
- A source link, if the model provided one, is marked "link not verified — check against the source".
Then write "Your conclusion" — what you concluded and what was prescribed — and click "Close case".
Two conditions without which the review will not start
- The materials are de-identified. No full names, dates of birth, document numbers, or telephone numbers. Check the form header separately — the name most often remains there.
- Consent to processing by an external model. The materials leave the platform, and this is a deliberate decision. The time of confirmation is recorded.
The same conditions apply to document recognition and image reading: without them the file will not be sent.
System prompts:
- if patient data is visible on a recognized form, the warning "Patient data visible — remove it from the text" will appear;
- in a DICOM file, personal data is often stored in service tags (name, ID, date of birth, address, telephone, institution, physicians) — it is not visible in the picture, but the system will list what it found. Only the slice image is sent to the model; tag values are not transmitted. It is better to upload a de-identified export.
Image reading: what is important to know
- Suitable for CT, MRI, radiography, ultrasound, echocardiography, ECG, EEG, ABR (BERA), endoscopy, and histology and cytology micrographs (a description of the frame, not a morphological report). First select the investigation type — otherwise the image will be examined "in general order", without a protocol.
- The result is a description of what is seen, not a report. The absence of findings on a single image does not mean the patient has none; the description always lists what limits the assessment.
- From a multi-frame DICOM, an evenly spaced sample of slices (up to 12) is read — a finding located between the displayed slices will not be noticed. Compressed or nonstandard DICOM is not read: the system will decline with an explanation.
- The accuracy of image reading has not yet been measured on a dedicated benchmark — this is why your verification and verdict are so important.
Storage and privacy
- Files are not stored. A photograph of a form, a PDF, and an image are used only for the duration of recognition; only the text enters the case.
- Case texts (question, context, materials, conclusions) are stored encrypted.
- A case is visible only to its author.
- Every submission of material to the model and every export is recorded in a protected access log before the data leaves the platform. More details —
/docs/privacyand/docs/hipaa.
Your verdict on each finding
Under each finding there is the question "Do you agree with this finding?" and three answers: Agree, Partially, Disagree. You can "Add a comment" — write what is correct (this is seen by the module's editor, not by the patient). The verdict can be changed at any time.
The verdict is not for reporting purposes: it shows where the review goes wrong, and it remains your annotation of the case.
Closing, reopening, exporting, deleting
- Close — only with your conclusion. In a closed case the materials cannot be changed.
- Reopen — for example, when further workup results arrive.
- Download the report — the case is exported as a single file (HTML), which can be printed or saved as a PDF from the browser. Your conclusion comes before the AI review. Findings you disagreed with are not removed from the document — that would be falsifying the history of the review.
- Delete a case — irreversibly, together with the materials, review, and conclusion.
How many reviews are available
Tasks are counted, not clicks: a case with radiography, CT, and laboratory tests launches three reviews and occupies three places in the quota. A repeat run ("Review again", "Try again") also counts.
By plan:
- Free (free physician plan) — the AI is disabled, no reviews.
- Trial period (the first 14 days after physician registration) — up to 100 reviews.
- Pro and Doctor Super — there is no separate review counter: each review deducts its cost from the plan's AI credits (300 credits per month on Pro, 1050 on Doctor Super), as with other AI functions. When credits run out, you can purchase an additional package or move to a higher plan. More details —
/docs/plans. - On former plans that are no longer sold, their own monthly review quota applies.
The month is counted as a rolling 30-day window, not a calendar month: places are freed up gradually.
General limits for all plans (protection against looping, not against spending):
- review — no more than 20 per hour and 60 per day;
- document and image recognition — no more than 40 per hour and 120 per day.
A refusal always states how long you must wait before retrying.
Other restrictions: file for recognition — up to 12 MB; PDF — up to 20 pages; up to 12 materials go into a single task; the laboratory panel — up to 120 values.
A draft CT report (see /docs/imaging-3d) consumes the same review quota.
Languages
The review and the image description are provided in the physician's interface language: Russian, English, Azerbaijani, Turkish, or Arabic.
How "Second Opinion" differs from neighboring sections
- AI Consultation (
/doctor/consultation-ai) — a dialogue with the AI on a medical question; "Second Opinion" is a structured review of the materials of a specific case with a protocol and verdicts. - Evidence-Based Medicine (
/doctor/evidence) — a search for evidence on a clinical question. - Simulators and the investigation library — educational, based on de-identified cases (
/docs/education,/docs/reference). A review of your own patient is not mixed with them. - AI patient summary from the chart — see
/docs/doctor-records.
Frequently asked questions
Why doesn't the "Review" button work? The two conditions (de-identification and consent) have not been confirmed, or the case contains no materials. If you are on the free plan — the AI is disabled there. If the limit has been exhausted, the message will indicate how long you must wait before retrying.
Can I upload a CT image or DICOM? Yes, via "Choose an image" (JPG, PNG, DICOM up to 12 MB). The AI will describe what it sees; from a series, a sample of up to 12 slices is read. Compressed DICOM is not read.
Are my files saved? No. The file is used only for recognition; the text remains in the case. Store the image or form itself in the patient's chart (/docs/doctor-records).
How much does one review cost? On Pro and Doctor Super — its actual cost is deducted in AI credits; the exact price depends on the volume of material. During the trial period — up to 100 reviews. On Free there are no reviews.
Why did one case "consume" several reviews? Each category (CT, laboratory tests, clinical case…) is reviewed as a separate task and counted separately.
Can the AI make a diagnosis or prescribe treatment? No. This is a supportive review; the conclusion is written and signed by the physician, and without your conclusion the case cannot be closed.
Can the patient see my cases and verdicts? No. A case is visible only to its author. Corrections to findings are seen only by the module's editor.
How do I obtain a document for printing? The "Download the report" button — the file opens in the browser and can be printed or saved as a PDF.
Can I return to a closed case? Yes, with the "Reopen" button — for example, when new results arrive.
Can I delete a case? Yes, the "Delete" button. Deletion is irreversible.
I am a resident — why can't I link a case to a clinic patient? Residents and students have access only to external and anonymous cases.
In which language will the review be provided? In your interface language: Russian, English, Azerbaijani, Turkish, or Arabic.
<!-- translated-from-ru: 97baed3e96dddd664fd4f599e191e9860485aeaa -->