Back to home

Pneumonia Detection

A CNN built to classify chest X-rays for pneumonia

Pneumonia Detection

Overview

A computer vision model that classifies chest X-rays into healthy or pneumonia-positive scans, built using transfer learning on VGG16.

The problem

Reviewing chest X-rays manually creates backlogs in busy clinics, which can delay diagnosis and treatment for patients with acute infections.

Notes

Fine-tuned a VGG16 backbone on chest radiograph data. Handled class imbalance with image augmentations (rotations, zooms, and horizontal flips) and learning rate scheduling to prevent overfitting.

Key Metrics

Reached 92.8% test accuracy and 94.2% recall on pneumonia cases, keeping false negatives low so active infections are rarely missed.

Tech Stack

Python, TensorFlow, VGG16, Deep Learning, NumPy, Pandas, Matplotlib, Seaborn

Next steps

Exporting to ONNX for lightweight drag-and-drop inference directly in the browser, then expanding the model to classify other pulmonary conditions.