Combining Focal loss with Cross-entropy loss for Pneumonia Classification with a Weighted Sampling Approach

Pneumonia, a severe respiratory infection affecting the lungs, stands as a leading cause of child mortality worldwide. The conventional method for detecting Pneumonia from chest X-rays relies on expert Pulmonologists manually identifying visual patterns, a time-consuming and specialist-dependent process. To address these limitations, this research introduces an automated Pneumonia detection system employing deep learning techniques with chest X-ray images. This study leverages the fine-tuned MobileNetV2 model for Pneumonia detection, incorporating a hybrid of two loss functions: cross-entropy and focal loss. Focal loss assigns greater importance to misclassifications within the minority class, while cross-entropy ensures that misclassifications in the majority class are adequately considered. To counteract class imbalance, random oversampling of minority class samples is applied. The proposed method's performance is rigorously evaluated using a publicly accessible dataset. Notably, the method achieves an impressive accuracy rate of 0.91 and an AUC (Area under the Curve) value of 0.97 for distinguishing between normal and Pneumonia-afflicted images. This demonstrates the robustness of the proposed approach in Pneumonia detection, harnessing the power of deep learning while effectively addressing class imbalance challenges prevalent in chest X-ray datasets.