# CT Analysis — Brain Bleed Detection

> A decision-support app that checks non-contrast head CT scans for bleeding inside the skull. A clinician drops in a DICOM folder, a zip file or a NIfTI volume. The app filters out scout images and removes patient…

- Client: Vela Neuro Clinic
- Year: 2026
- Category: Apps & Software

## Overview
A decision-support app that checks non-contrast head CT scans for bleeding inside the skull. A clinician drops in a DICOM folder, a zip file or a NIfTI volume. The app filters out scout images and removes patient identifiers. In about 2.5 seconds on a laptop GPU, it reports the probability of any bleed and of each of five bleed types, with a per-slice probability curve and heatmaps showing where it found blood. Every result is marked Detected, Indeterminate or Not detected, and the app warns when a scan falls outside what the model was built for, such as contrast scans or children. Radiologists record their own read on a form that starts blank, so the model's output doesn't sway them, and those reads feed a live performance page and the calibration tools. Built with FastAPI, PyTorch and React around an open, MIT-licensed model trained on the RSNA brain haemorrhage dataset. For research use only: it is not a certified medical device, and the screenshots show synthetic scans.

## Deliverables
- Upload of DICOM folders, zip files and NIfTI volumes, with automatic removal of patient identifiers
- Whole-scan and per-slice bleed probabilities for five bleed types
- Slice viewer with CT window presets, heatmaps and keyboard navigation
- Warnings for contrast, paediatric, reformatted or incomplete scans
- Clinician review form and a live model-performance page
- Calibration and external-validation scripts, plus a roadmap to medical-device certification

Start a similar project: https://graphtsy.es/contact
