Auto-MyIn: Automatic diagnosis of myocardial infarction via multiple GLCMs, CNNs, and SVMs

This paper proposes an automated diagnostic tool namely, Auto-MyIn, for diagnosing myocardial infarction (MI) using multiple convolutional neural networks (CNN). Rather than utilizing the spatial information of the original delayed-enhancement magnetic resonance (DE-MRI) images, Auto-MyIn uses the textural information obtained by applying grey level co-occurrence matrix (GLCM) of four different grey levels to train the three CNNs (ResNet-18, DarkNet-19, and SqueezeNet). First, the images generated from each GLCM grey level are used to train each CNN individually. Next, for each GLCM grey level, the textural-based deep features extracted from the three CNNs are concatenated and used to train several support vector machine (SVM) classifiers. Finally, Auto-MyIn fuses textural-based deep features of the four GLCM grey levels using principal component analysis (PCA). The results of Auto-MyIn indicated that fusing the textural-based deep features of each level of GLCM is better than the end-to-end deep learning classification of the three CNNs trained with each grey level of GLCM images. Furthermore, it showed that fusing textural-based deep features of the four grey levels of the GLCM using PCA has further improvement on diagnostic accuracy. Moreover, the results prove that using textural information is superior to using spatial information of the original DE-MRI images. In addition, the results of Auto-MyIn when compared with other related studies demonstrated its competitive ability. Moreover, the performance of Auto-MyIn shows an accuracy of 0.984, a sensitivity of 0.992, specificity of 0.968, and precision of 0.967, which indicate that it is a reliable tool, therefore it could be employed to help in the clinical decision making and facilitate the diagnostic process of MI thus avoiding the limitations of manual diagnosis.

Automated summary

This paper proposes an automated diagnostic tool, Auto-MyIn, for diagnosing myocardial infarction (MI) using multiple convolutional neural networks (CNN). The tool utilizes textural information obtained from grey level co-occurrence matrix (GLCM) and principal component analysis (PCA) to improve diagnostic accuracy. The results show that fusing textural-based deep features and using textural information is superior to using spatial information of the original DE-MRI images. The performance of Auto-MyIn indicates its reliability and competitive ability compared to other related studies.