OncoVision API

OncoVision API is a deep learning–based medical image analysis system that detects cancer from medical images (histopathology / X-ray / MRI depending on dataset) using a Convolutional Neural Network (CNN).
OncoVision API is a backend-only academic project that performs cancer detection from medical images using a Convolutional Neural Network (CNN).

The system accepts file uploads of multiple types, safely converts them into CNN-compatible images, and performs inference using a pretrained model.

uvicorn app.main:app --reload

  --------------------------------------------------
  Key Characteristics
  --------------------------------------------------
  - Backend only (FastAPI) 
  - CNN-based inference (no transformers, no LLMs) 
  - Fully offline (no cloud APIs) 
  - File-based input only (multipart upload) 
  - Supports images and documents 
  - Beginner-friendly but architecturally clean 
  - Academic + production-aligned design
  --------------------------------------------------

Architecture Overview

Client -> FastAPI Route -> Inference Service -> File Converter -> Image Preprocessing -> CNN Prediction -> Structured JSON Response

Core Design Principle: The CNN only works on images. All non-image files are converted into images before inference.

  --------------------------------------------------
  Supported Input Types
  --------------------------------------------------
  Images: JPG, JPEG, PNG Documents: PDF, DOCX
  Others: Rejected
  --------------------------------------------------

API Response Format

{
 “status”: “success”,
 “file_type”: “pdf”,
 “images_processed”: 3,
 “prediction”: “Cancer Detected”,
 “confidence”: 0.87
}

OncoVision-API/
├── backend/
│   ├── __init__.py
│   ├── app/
│   │   ├── __init__.py
│   │   ├── api/
│   │   ├── model/
│   │   ├── services/
│   │   ├── utils/
│   │   ├── config.py
│   │   └── main.py
│   ├── model_weights/
│   └── temp_uploads/
│
├── venv/
├── README.md
└── requirements.txt

Academic Disclaimer

This project is for educational purposes only. It is not a medical device and must not be used for real-world diagnosis.