Custom CNN Model for Pheumonia Detection using Chest XRays

This project was a part of my 7th Semester Project as part of Engineering Course

Project Overview:

The project focuses on developing a lightweight and efficient Convolutional Neural Network (CNN) for pneumonia detection using chest X-ray images. I benchmarked well-known architectures like MobileNetV2, ResNet152V2, and Xception, and explored attention mechanisms (SE & CBAM) to enhance feature representation.

Key Highlights:

• Designed a Custom CNN model from scratch that achieved ~93% accuracy with minimal memory (0.92 MB), far more efficient than heavy pre-trained networks.
• Integrated Squeeze-and-Excitation [SE] and Convolutional Block Attention Mechanism [CBAM] attention to test the cross-domain generalization under domain shift conditions.
• Conducted extensive analysis of FLOPs, MACs, and model complexity, balancing high diagnostic accuracy with real-world deployability.
• Tested the model on an external dataset to ensure robustness and adaptability.

This work not only strengthened my foundations in Deep Learning and Medical Image Analysis, but also helped me appreciate the trade-offs between accuracy and computational efficiency; especially for edge AI in healthcare.

Project Collaborators:

• Guide Prof. Ralph Samuel Thangaraj
• Niket Girdhar (me)

Detailed Project Report: