The chest X-ray stands as a commonly employed radiological examination for identifying thoracic ailments. Despite the advancements brought by convolutional neural network (CNN) techniques in categorizing thoracic conditions through these images, the varying scales of pathological irregularities across different thoracic diseases present an ongoing challenge. In response to these concerns, this study proposes a refinement to the VGG19 model, a well-known residual network architecture. This enhancement involves integrating a pyramidal convolution module and a shuffle attention module, both addressing the previously mentioned issues. Specifically, the introduced VGG19 model leverages the shuffle attention mechanism to focus on distinctive traits of pathological abnormalities. This mechanism augments the capacity of the pyramid convolution component, allowing it to extract more discerning features related to pathological irregularities compared to the conventional 3x3 convolution. Rigorous assessments conducted on the ChestX-ray14 and COVIDx datasets underscore the VGG19 model's superior performance over other advanced methodologies. Furthermore, an ablation study is carried out to delve deeper into the impact of pyramidal convolution and shuffle attention on enhancing the classification efficacy of thoracic diseases. The study results bolster the evidence indicating the effectiveness of these integrated components in augmenting the model's proficiency in thoracic disease classification.

Keywords : VGG19, CXR (Chest X-ray), Consult Net, Pyramidal Convolution Module and Shuffle Attention Network (PCMSANet), Relu.