Pneumonia Detection with Deep Neural Networks

Comparing four architectures for pneumonia detection from chest X-rays.

Python
TensorFlow
Keras
DenseNet121
ResNet50
EfficientNet-B3
Streamlit

Conceptual illustration created for this case study — not a screenshot of the application.

Overview and context

An academic project at Memorial University comparing deep-learning architectures for automated pneumonia detection from chest X-ray images. This is a coursework and research exercise: the results below are project evaluation figures on the project dataset, not evidence of clinical validation or diagnostic readiness.

Problem

Chest X-ray classification is a common benchmark, but results depend heavily on class imbalance handling, overfitting control, and decision thresholds.

My role and contribution

Designed and evaluated the model set: DenseNet121, ResNet50, EfficientNet-B3, and a custom CNN.
Addressed class imbalance, overfitting, and model reliability through systematic evaluation.
Optimized the decision threshold rather than accepting a default cut-off.
Deployed the selected model through a Streamlit interface.

Technical approach

Four architectures trained and compared on the same chest X-ray dataset.
Class imbalance and overfitting treated as first-order problems in the evaluation design.
Decision-threshold optimization used to balance precision and recall.
A Streamlit interface used to make the model interactively testable.

Evidence and outcomes

Reported project results: 88.2% accuracy, 90.4% precision, 90.3% recall.
A comparative view of how the four architectures behaved on the same task.