ResearchHealthcare

Segmenting and classifying tuberculosis, pneumonia and COVID-19 in chest X-rays

A deep learning pipeline that first isolates the lungs in a chest X-ray, then classifies the image as normal, tuberculosis, pneumonia or COVID-19.

The challenge

Chest X-rays are the most widely available imaging test in India, but reading them takes expertise that is scarce in many districts. Tuberculosis, pneumonia and COVID-19 can look similar, and image artefacts outside the lungs can mislead automated tools.

Our approach

Segment the lung fields so that later steps focus only on lung tissue.
Train deep learning classifiers on the segmented lungs to separate normal, TB, pneumonia and COVID-19 images.
Compare architectures and evaluate on held-out data to measure how well the approach generalises.

The solution

Lung segmentation model
Multi-class disease classifier
Evaluation on held-out test images

Results

Peer-reviewed and published in Multimedia Tools and Applications (Springer), 2025.
The segment-then-classify approach informs the imaging module of BigHealth.