Objectives: Pneumonia is the third most common cause of death in Singapore. Early detection of pneumonia is the key for prevention but diagnosis requires an accurate classification of chest X-ray images, which can often have a ambiguity and variability. This project aims to leverage on CNN with the goal of ultimately supporting in the classification of chest X-rays for quick, timely and accurate pneumonia detection.
Dataset: Chest X-ray dataset provided by Doctor Anywhere
Models Explored: 1) Model Alpha - customized simple CNN architecture built from scratch, loosely based on VGG architecture, 2) Model Beta - a pre-trained DenseNet 121 model
Best Results: Accuracy of 98% with FNR of 1.4% (training time of ~2 mins with T4 GPU)