| Summary: | Magnetic Resonance Imaging (MRI) typically shows the overall heart anatomy and usually includes the outmost slices of the left ventricle coverage. In assessing the patient in the left ventricle (LV) cardiac segment, only slices with images of the LV cardiac segment are considered, and the rest is neglected. This chapter explores an automated approach to classifying LV and Non-LV segments in cardiac MR images by utilizing a deep convolutional neural network. The dataset used is the STACOM2012 public dataset, which consists of 398 short-axis images of cardiac LGE-MRI. A deep convolution network model designed from scratch and three deep transfer learning models (AlexNet, GoogleNet and SqueezeNet) were trained on 80% of the images and validated on the remaining 20% of the images after the data augmentation process for a comparative analysis using three different optimization algorithms (ADAM, SGDM and RMSprop). Then, all networks were tested on cardiac LGE-MRI collected from Advanced Medical and Dental Institute (AMDI) USM database. The outcome demonstrated that Adam was the best network optimizer, with an accuracy improvement of 0.3-1.1% over SGDM. The GoogleNet model outperformed other models with an accuracy performance of 97.54% and a macro F1-score of 0.9080 when tested with the STACOM2012 and AMDI datasets, respectively. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.
|
|---|