~/projects/pneumonia-detector2022Deep LearningTensorFlowPyQt
Pneumonia Detector using Deep Learning
Chest X-ray diagnosis with transfer learning
pneumonia-detector.proj
Role
ML Engineer
Timeline
3 months
Year
2022
Type
Project
// Key Results
90%
Accuracy
<1s
Inference Time
5.8K
Training Images
01 // The Problem
What was broken
Radiologist-led diagnosis of pneumonia from X-rays is slow, especially in under-resourced clinics. There was a need for a screening tool accessible to non-specialists.
02 // The Approach
How I thought about it
Fine-tuned pre-trained convolutional networks (VGG16, ResNet50) on a labeled chest X-ray dataset. Wrapped the inference pipeline in a PyQt desktop app with speech I/O.
03 // The Solution
What I built
A cross-platform desktop application that accepts an X-ray, returns a pneumonia probability with Grad-CAM heatmap, and announces results via voice for accessibility.
04 // The Impact
Outcomes & learnings
Reached 90% classification accuracy on a held-out test set, with sub-second inference on CPU — suitable for deployment in low-resource clinical environments.
// Tech Stack
PythonTensorFlowKerasPyQt5OpenCVSpeech API