Report review
Source reports, model responses and saved clinical evidence keep their original language.
For clinicians, researchers, educators and students: choose a model, add report text, then compare each suggestion with its source. This public preview uses synthetic reports only.
1. Choose your model
Choose one connection. Then add a complete report and check the source evidence.
Connect directly to Ollama you control, with cloud features disabled. The selected model is checked before any report text is sent. No cloud fallback is used.
On shared servers, ask the administrator before changing Ollama. Local connection and SSH tunnel instructions.
The local address refers to the device running this browser. A phone or tablet needs its own approved connection to your model server. Your browser may request local-network permission.
Use your own API key
Already configured? Enter your gateway access code below and connect. First time? Start the personal gateway once using the setup guide.
First-time API setup
One-time setup on your computer:
- From your ClinLoop installation, run
python scripts/run_research_review.py --setup. - Open Personal API setup. Add your provider key and exact model ID there. Copy the separate gateway access code from the private file shown in your terminal.
- Select the same provider below, paste the gateway access code and connect.
The setup link refers to this device. For a gateway on another computer, use the private tunnel instructions. Provider model access and API billing are separate from this website. No model is guaranteed to be accurate for every health report.
External transmission: synthetic text will go from your local gateway to OpenAI. Provider terms and retention policies apply. No real patient information. Provider API keys stay in your personal gateway and must not be entered on this public page. Use Personal API setup on your computer or the terminal helper. Changing provider clears the report and requires a new connection and confirmation; no provider fallback is used.
Disconnected. No report has been sent.
2. Load and check a report
You can load, preview and check a synthetic report before connecting a model. Only Review report sends text to the selected model.
Local file tools for PDF, Word DOCX, scans, photos and DICOM
Start your updated personal gateway on this computer: python scripts/run_research_review.py --setup
No provider API key is needed to import files. Install the pinned API dependencies and Tesseract with the selected OCR language. Files go only to this loopback gateway. OCR may create temporary files on your computer; no report history is kept.
File formats and limits
One file up to 5 MB; PDF/images up to 20 pages/frames. TXT, CSV, TSV and Markdown must be UTF-8 and fit 6000 bytes. Complete extracted documents can be previewed up to 120000 bytes, but model review requires the whole report to fit 6000 bytes. Nothing is silently shortened.
TXT, CSV, TSV and Markdown load as original text in this browser; formulas and markup are never executed. PDF, plain Word DOCX, scanned reports, PNG/JPEG/TIFF photos and DICOM use your personal local gateway. Complex Word documents, legacy DOC, spreadsheets and proprietary formats require a complete PDF or UTF-8 text export. Image pixels and waveforms need a separate validated workflow.
For real-patient use and imaging interpretation, use your institution’s approved installation. The document/imaging workflow overview on this public site is a synthetic demo. Check extraction against every original page before interpretation.
You can load and check a report now. Connect a model before selecting Review report.
3. Compare each suggestion with the report
Each suggestion includes nearby source text so you can check negation, dates and context. Repeated quotes are marked as ambiguous. Source matching proves that quoted words exist, not that a model’s interpretation is correct. Missing or empty results never mean normal findings or completed care.
Session privacy
This page keeps no saved report history and uses no analytics. Report text, results and the gateway access code stay in page memory; disconnecting, changing pathway, leaving or 10 minutes of inactivity clears them. Clearing does not erase model/provider logs, revoke API credentials or guarantee secure deletion from device memory.
The public website is not a hospital patient database. Hospital care must use an approved on-premises installation with local LLMs and its normal authentication, access recording and audit trail. The research API is a separate synthetic-only service with no connection to hospital records.
Plan an approved report study승인된 보고서 연구 계획
Use the offline study coordinator to freeze source reports and model outputs, prepare separate blinded reference worksheets, and measure extraction agreement after independent review. It supports the six text categories above; it does not analyze image pixels or establish diagnostic accuracy.오프라인 연구 도구로 원문과 모델 출력을 고정하고, 예측을 표시하지 않는 별도 참조 검토지를 준비한 뒤 독립 검토 후 추출 일치도를 측정하세요. 위의 여섯 가지 텍스트 유형을 지원하며 영상 화소 분석이나 진단 정확도 입증은 제공하지 않습니다.
For real-data studies, the institution must authorize the private environment, data use, ethics arrangements and clinical oversight. Students use synthetic examples or institution-approved study access. Naming a study site does not establish approval.실제 데이터 연구에는 기관의 비공개 환경, 데이터 사용, 연구윤리 절차 및 임상 감독 승인이 필요합니다. 학생은 합성 예제 또는 기관이 승인한 연구 접근 권한을 사용하세요. 연구 기관을 지정하는 것만으로 승인이 성립하지 않습니다.
Offline study instructions and synthetic example오프라인 연구 안내 및 합성 예제