CLASSIFICATION OF CARDIOMEGALI BASED INTELLIGENT ALGORITHM CONVOLUTIONAL NEURAL NETWORKS

• Published in: Journal of medical imaging and radiation sciences Vol. 53; no. 4; p. S56
• Main Authors: Pradana, Datu Euphratadhi, Sudiyono, Rochmayanti, Dwi
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.12.2022
• Subjects: Cardiomegaly; deep learning; neural network convolution; Cardiomegaly; deep learning; neural network convolution
• ISSN: 1939-8654; 1876-7982
• DOI: 10.1016/j.jmir.2022.10.183

Cardiomegaly is known as an enlarged heart. From the thoracic radiographic image, radiologists use cardiothoracic ratio (CTR) as a measurement scale to determine abnormal heart size or indications of cardiomegaly (1). The accuracy of using CTR to detect an increase in heart size and predict cardiomegaly is 95.8% (2). Manually measuring CTR requires significant work and takes time. Recent advances in artificial intelligence with deep learning have significantly increased performance in computer assisted cardiomegaly detection to a level comparable to that of radiologists (3). One type of deep learning algorithm that is used to identify cardiomegaly is the convolutional neural network (CNN). CNN has a good performance on information processing and visual form recognition (4). The research is to develop intelligent algorithm design for cardiomegaly classification. Knowing the accuracy of sensitivity, specificity, positive predictive value and negative predictive value of the CNN-based deep learning program in classifying the heart into cardiomegaly or normal heart. Research and development with post test only control group design on chest radiography images. Build deep learning applications through python programs. The thoracic image used as data set was 1258, training data used 1177 antero-posterior and postero-anterior projection thoracic images, the test data used 81 postero-anterior projection thoracic images. Tests were carried out by measuring accuracy, sensitivity, specificity, positive predictive value, negative predictive value and curve receiver operating characteristics. Testing of deep learning applications yields value accuracy is 86.41%, sensitivity of 95%, specificity 78.05%, the positive predictive value is 80.85%, negative predictive valueas is 94.11% dan value of area under curve (AUC) 0.875. Based on the results of analysis and research, deep learning applications based on convolutional neural networks, can be used as an alternative reference for heart classification into cardiomegaly or normal heart.