OncoDetect

Making a model usable is a software problem, too.

23.85Mmodel parameters

OncoDetect brings image intake, model readiness, preprocessing, inference and saved observations into one application. The focus is the complete path from a validated upload to a traceable result.

Implementation & scope

• Integrated a ResNet50-based model with 23,850,113 parameters and 224 × 224 RGB preprocessing.
• Connected a React interface to FastAPI and SQLite-backed observation history.
• Added 14 API and interface regression tests for uploads, errors, history and workspace reset.

Reading the results

This is a research and software portfolio project. Model size and passing software tests do not establish clinical accuracy; it is not a diagnostic service.