DEVELOPMENT OF CONVOLUTIONAL NEURAL NETWORK MODEL FOR CLASSIFICATION OF CARDIOMEGALY X-RAY IMAGES

Abstract

In this study, the effect of the input image format on the accuracy in deep learning (DL) was evaluated using tagged image file format (TIFF), PNG, and JPEG images converted from DICOM images. Based on the evaluation results, a convolutional neural network (CNN)-based DL model was proposed. The training and test data used in this study were chest X-ray images of patients diagnosed with normal heart or cardiomegaly conditions. CNN models, namely VGGNet, ResNet, InceptionNet (GoogleNet), DenseNet, and EfficientNet, were used to compare the results according to the types of input images, that is, images converted from DICOM images into TIFF, PNG, and JPEG images. The classification performance was validated by proposing a CNN model that can classify normal hearts and cardiomegaly in chest X-ray images. In this study, through medical imaging research using deep learning, it was demonstrated that the classification performance was unaffected even when the DICOM image was converted into any format and used as an input image. In addition, the proposed CNN model exhibited excellent performance in classifying normal hearts and cardiomegaly on X-ray images. This can be used in various studies that aim to apply DL to medical images by providing information according to the input image type and is also considered to aid in the selection of image type and learning parameters. The proposed model is expected to yield useful results for the classification of diseases in chest X-ray images.