🎯 The Objective

To assist radio-oncologists in quickly identifying probable cancerous lumps in Contrast-Enhanced Spectral Mammography (CESM) images without immediate need for stat biopsies.

Figure 1: Workflow.

Figure 1: Workflow.

🧠 Approach

  • Data Processing: Utilized subtracted images to isolate potential masses.
  • Architecture: Implemented Transfer Learning on shallow neural networks to maintain efficiency.
  • Explainability: Integrated Saliency Heatmapping and GradCam to visualize why the model made a prediction—crucial for medical validation.

🏆 Results

Preliminary performance metrics indicate the model outperforms standard radiologist benchmarks in specific classification tasks. The results were validated in collaboration with a radio-oncologist.

📄 Read Full Research Report (PDF)