10Deep Learning
Lung Disease Detection
A chest X-ray analysis prototype that uses deep learning to identify potential pneumonia regions and display annotated results.
Project preview01 / 01
The idea behind it
Lung Disease Detection connects a React interface to a Django backend for chest X-ray analysis. A TensorFlow/Keras segmentation model processes 256x256 grayscale images, and the backend turns predicted regions into bounding boxes. The interface displays the uploaded X-ray, the annotated result, a normal-or-pneumonia prediction, and a model confidence score.
The repository documents a U-Net architecture and a combined intersection-over-union and binary cross-entropy loss. This project demonstrates an end-to-end medical imaging workflow from image upload to visual model output.
What it does
- Chest X-ray upload and preview in a React interface
- TensorFlow/Keras segmentation on 256x256 grayscale images
- Predicted-region extraction using connected components
- Bounding-box annotations over the original X-ray
- Normal-or-pneumonia output with a model confidence score
- Django upload endpoint connecting the interface to model inference
Challenges & solutions
- Preparing grayscale images and masks at a consistent model input size
- Turning segmentation probabilities into visual regions and bounding boxes
- Connecting image upload, backend inference, and result rendering
- Combining overlap-based and binary classification loss functions
Outcomes & impact
- Implemented an image-upload-to-segmentation workflow across React and Django
- Presented annotated images alongside classification and confidence outputs
- Documented the model workflow, API, and application architecture