An Experimental Analysis on CNN and Transfer Learning Models for Pneumonia Detection Using Chest Radiographs

Pulmonary infectious diseases, such as pneumonia, pose a significant public health concern. Accurate diagnosis is crucial for ensuring effective treatment and improved patient outcomes. Digital biomedical records, such as chest X-ray imaging, are crucial diagnostic tools for detecting respiratory disorders. The complexity and variability of these images make it challenging for even skilled radiologists to analyze them. This is where technology plays a vital role. Here, we explore the use of convolutional neural networks (CNNs), a powerful deep learning approach, to efficiently classify viral respiratory diseases. Chest X-ray scans can enhance diagnostic accuracy and efficiency by detecting signs of disease progression, such as pneumonia. The analysis involves the use of 17,621 chest X-ray scans that are classified as either normal or pneumonia. This dataset is used for adapting and testing various pre-trained CNN models, such as DenseNet, VGG16, ResNet and InceptionNet along with other models using classification metrics. We have used a custom CNN based method to differentiate between pneumonia and normal cases, which is more accurate than transfer learning models. This approach can be advantageous for radiologists, particularly in the resource limited settings, CNNs’ potential is highlighted more in the medical image classification.