Lung Disease Detection: Comparing Sequential, Functional and VGG-16 CNNs

Overview

Undergraduate research paper co-authored with Rohit Chauhan and Suryajeet Gupta at Quantum University. We benchmarked three CNN variants — a custom sequential model, a custom functional (multi-branch) model, and a VGG-16 transfer-learning setup across chest X-ray and CT-scan classification for three conditions.

Results

• Pneumonia (sequential, chest X-ray): F1 98.55%, accuracy 98.43%, recall 96.33%
• Tuberculosis (sequential, chest X-ray): F1 97.99%, accuracy 99.4%, recall 98.88%
• Lung cancer (functional, CT scans): accuracy 99.9%, specificity 99.89%

Sequential model outperformed on X-ray tasks; the functional model (7×7 and 1×1 branches concatenated, feeding five 3×3 conv layers) generalized better on CT cancer classification.

Stack

• Framework: TensorFlow / Keras
• Datasets: Covid-19 radiography database (Kaggle), Mendeley chest X-ray posteroanterior collection, lung CT-scan dataset
• Preprocessing: 224×224 resize; horizontal flip, zoom, shear, rotation, rescale augmentation
• Sequential: 5 conv layers, LeakyReLU (α=0.66), max pooling, Adam @ lr=1e-4, 50 epochs
• Functional: 7×7 and 1×1 branches over 3×3, concatenated, then five 3×3 conv layers
• Pretrained: VGG-16 with ImageNet weights

What I’d change next

The reported accuracies are strong but the test sets are small (85 images for TB, 278 for cancer) and the training data leaves external validity unproven. A properly held-out external cohort, cross-validation, and interpretability checks would be table stakes today — themes that motivated my later work.

Links

• Preprint (unpublished): PDF
• Notebook: github.com/aryanp2107/DL-Projects