Reshape
Far far away, behind the word mountains, far from the countries.
Chest X-ray Pneumonia & Heart Analysis
This project applies state-of-the-art deep learning techniques to automate detection of pneumonia and quantitative assessment of cardiomegaly from chest X-rays:
Architectures Evaluated
- Baseline CNN
- Enhanced/custom CNN
- VGG-15 (pretrained)
- ResNet-142 (pretrained)
Heatmap Generation
- Leveraged Grad-CAM (tf-explain) to produce class activation maps
- Pinpointed regions of lung consolidation (pneumonia)
- Highlighted cardiac silhouette for heart size measurement
Data Curation & Augmentation
- Combined 512 GB Stanford Chest X-ray collection with 42 GB Kaggle dataset
- Automated augmentation (rotations, flips, intensity shifts)
- Filtered to yield a balanced 2.8 GB training corpus
Results & Deployment
- Enhanced CNN achieved R² score of 96% on held-out data
- Packaged as an interactive Streamlit web app for on-demand image upload, inference, and heatmap visualization
- Methodology formalized in a published concept paper
End-to-End Pipeline
This pipeline—from raw DICOM ingestion through model inference to web-based visualization—demonstrates both high diagnostic accuracy and practical usability for clinical decision support.
“
"This end-to-end solution provides clinicians with rapid, interpretable, and highly accurate diagnostic support for pneumonia and heart size assessment from chest X-rays."
