Abstract

Study a detailed method for classifying and studying medical images using sets of data from the NIH Chest X-ray Collection and the VinDr-CXR collection. There are 51,759 examples in these data, and each one includes 15 categories explaining various problems found in chest X-ray images. More than 15,000 images are included in the 14 subcategories about lung diseases. The goal is to use the latest deep learning methods to allow machines to identify and sort various disorders of the chest. In pre-processing stage, tasks such as encoding images, improving learning through generalization, and designing the TensorFlow dataset are handled. The data is put into training and testing sets and illustrated with Matplotlib. A DWT is applied to the images to help reduce their noise. The features these frameworks, Seresnet152 and ResNet, help collect and use for image classification are known and appreciated. They are made by combining Keras and TensorFlow to stack a pre-trained model with extra layers. Experts pay close attention to the loss function, the activation function, and the total number of parameters. The model’s accuracy, Roc_Auc, and loss values demonstrate that the model is performing successfully. Getting 90.26% accuracy, Roc_Auc of 92.75% and having low values for loss is what was achieved. The outcomes indicate that the two models have the potential to help doctors diagnose patients with chest X-rays. The findings point out that medical images are analysed by deep learning models that are based on correct data, appropriate model design, and in-depth tests. Because of all these factors, the model can handle real-case challenges successfully.