Robust stochastic gradient descent with momentum based framework for enhanced chest X-ray image diagnosis

Our study focuses on the difficulties that radiologists encounter while interpreting Chest X-Ray images, which arise from the inherent ambiguity. We propose an innovative algorithm that outperforms earlier techniques. We employ deep convolutional networks and a pre-trained DenseNet with 121 layers to identify fourteen diseases, such as Pneumonia, Edema, and Emphysema. This approach enhances both the capacity and interpretability of the model. By implementing a strong dataset splitting technique and utilizing Stochastic Gradient Descent with momentum, our research showcases exceptional results and outperforms previous optimizers. The model not only surpasses the performance of ordinary radiologists in the F1 metric, but also demonstrates significant enhancement in Area Under Curve (AUC) scores. This is a notable progress in the automated detection of diseases in Chest X-Rays, offering improved accuracy in diagnosis and wider ramifications for the field of radiography and patient treatment. The source code, pre-trained model and dataset are available at https://github.com/debajyoty/Chest_X-Ray_Diagnosis-main.git.