Cardiomegaly Disease Classification Using Convolutional Neural Networks and EfficientNet

Abstract

The deep learning approach for automatically identifying cardiomegaly from chest X-ray images using Convolutional Neural Networks (CNN) and EfficientNet architectures is described in this article. The collection, which was sourced from the "Chest X-ray" repository, includes a variety of labelled chest X-ray images arranged into test, validation, and training sets. A strong basis for model training and assessment is established by annotating each image to indicate whether cardiomegaly is present or not. To balance depth, width, and resolution, the EfficientNetB7 model—which had previously been trained on ImageNet—was adjusted for cardiomegaly classification using its compound scaling. A CNN model based on VGG16 was also employed for comparison. To increase generalizability, both models were trained with data augmentation techniques. Training, validation, and testing accuracy for the EfficientNet model were 98%, 96%, and 95%, respectively. CNN, on the other hand, achieved 91% test accuracy, 92% validation accuracy, and 94% training accuracy. The categorization report, confusion matrix, accuracy, and loss were the evaluation measures. The results show that EfficientNet has potential for medical image processing, offering a dependable and effective way to identify cardiomegaly. This study highlights how to improve diagnostic speed and accuracy in medical imaging by utilizing sophisticated deep learning models.