Pneumonia-Detector

Icon
Landing Page (LP-1)
Disclamier and LP-2
Main Workspace-3 (Patient Histroy)
Main Workspace-2(Final Analysis)
Main Workspace-1( Overview dashboard)
PDF- Report

Inspiration

Pneumonia remains a leading cause of mortality worldwide, especially in areas with limited access to expert radiologists. We wanted to build a fast, reliable, and accessible diagnostic tool that assists healthcare workers in detecting pneumonia from chest X-rays quickly.

What it does

X-Ray Analysis: Accepts chest X-ray uploads and accurately classifies them as normal or pneumonia-positive.
Explainable AI (Grad-CAM): Highlights specific lung regions influencing the model's decision with visual heatmaps.
Clinical Reporting: Generates instant diagnostic summaries with confidence scores for clinical evaluation.

How we built it

Deep Learning & Computer Vision: PyTorch, Torchvision, OpenCV, and Grad-CAM for image classification and visual explanation.
Backend: FastAPI and Uvicorn for lightweight, high-performance API endpoints.
Frontend: Next.js, React, and Tailwind CSS for a clean, responsive medical console.
Deployment & Cloud: Vercel for frontend hosting and Railway for cloud model inference.

Challenges we ran into

Cloud Memory Constraints: Handling PyTorch and Grad-CAM memory overhead on cloud servers without triggering Out-of-Memory (OOM) errors.
Cross-Origin Configuration: Resolving CORS policies and environment routing between the decoupled Vercel frontend and cloud backend.
Model Explainability: Accurately mapping convolutional layer activations back onto variable-sized medical input images.

Accomplishments that we're proud of

Successfully deployed an end-to-end deep learning pipeline connecting a modern web UI to a live inference backend.
Integrated Grad-CAM heatmaps to ensure the model provides transparent, interpretable visual evidence rather than acting as a black box.
Maintained low latency for real-time diagnostic predictions.

What we learned

Best practices for containerizing and optimizing heavy ML dependencies (PyTorch/OpenCV) for production environments.
The critical importance of explainable AI in medical imaging to build clinical trust.
Seamlessly managing cross-platform deployments between Vercel and cloud compute providers.

What's next for Pneumonia-Detector

Multi-Class Detection: Expanding the model to identify COVID-19, tuberculosis, and other respiratory conditions.
DICOM Support: Enabling direct ingestion of standard medical imaging formats used in hospital PACS systems.
Mobile Accessibility: Optimizing the interface and lightweight inference for mobile field clinics.