AI to detect and classify Pneumonia from Chest XRay Images
The goal of this project is to buit a deep learning model that can diagnose Pneumonia from Chest XRay Images. The dataset is available for download at https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia. In this classification problem, "Recall" is the most significant performance metric even more than accuracy and precision. False negative has to be minimized because falsely diagnosing a patient of pneumonia as not having pneumonia is a much larger concern than falsely diagnosing a healthy person as a pneumonia patient. By minimizing false negative, which is in the denominator, we can increase 'Recall'. I have built a Convolutional Neural Network (CNN) model that achieves a recall of 99%. I have also implemented a DenseNet121 model similar to the CheXnet model (CheXNet https://stanfordmlgroup.github.io/projects/chexnet/) and used the weights of the CheXNet model available at https://github.com/brucechou1983/CheXNet-Keras. This model achieves a Recall of only 43%. Both the models were trained for 5 epochs.