Performance Analysis of Residual Networks for Pneumonia Diagnosis on Chest X-Rays

ResNets have demonstrated exceptional capabilities in image classification tasks, making them a promising tool for clinical applications. This study evaluates the performance of deep learning Residual Networks (ResNets) in classifying chest X-ray images for pneumonia diagnosis. Five ResNet variants: ResNet18, ResNet34, ResNet50, ResNet101, and ResNet152 are compared on their performance training on the same dataset of chest X-ray images, which included positive and negative diagnoses. The models were trained on a subset of images and tested on a separate set to assess their accuracy and consistency. Our results indicate that ResNet50 achieves the highest accuracy in predictions, while ResNet34 demonstrates the most consistent performance across trials. These findings provide valuable insights into selecting the most suitable ResNet variant for binary classification tasks involving chest X-rays, particularly when considering factors such as computational resource availability and accuracy requirements.