BreatheWish
AI-assisted chest radiograph screening and clinical triage platform featuring a DenseNet-121 classification backbone, Grad-CAM visual explainability, and a decoupled FastAPI clinical workflow service.
Emergency radiograph queues cause diagnostic bottlenecks for acute pulmonary consolidation. BreatheWish provides instant preliminary screening and interpretable saliency heatmaps to assist clinicians.
Achieved 90.89% test accuracy and 0.9906 ROC-AUC with verified Grad-CAM saliency maps for clinical audibility.
End-to-end execution pipeline
Document normalization, raw text extraction, and domain chunking
Dense vector representations via fastembed and FAISS vector indices
Top-k semantic similarity matching with thresholded relevance verification
Constrained LLM context injection with strict source citation formatting
Core technologies & frameworks
Open-source clinical prototype demonstrating interpretable deep learning triage with fail-closed safety contracts.
Integrated automated Grad-CAM explainability and asynchronous physician-patient communication channels.
Key engineering learnings
Trained and evaluated a DenseNet-121 architecture achieving 98.25% specificity on normal radiographs.
Engineered real-time Grad-CAM saliency heatmaps overlaid on medical DICOM/radiograph imagery.
Built a fail-closed API architecture with role-based access control across doctor and patient workflows.