Optimal Deep Learning Based Atherosclerotic Plaque Classification on Intravascular Ultrasound Images

Abstract

Intravascular ultrasound (IVUS) is suitable for evaluating plaque and lesion morphology features and helps to make clinical decisions during the treatment of coronary artery disease (CAD). IVUS remain a gold standard in accessing atherosclerosis plaques, coronary lesions, and stenosis. Even though plaque classification by IVUS is essential for risk stratification, frame-by-frame analysis of an entire vascular segment is labor intensive. In recent times, in the field of deep learning (DL), convolutional neural network (CNN) has been intended to adaptively and automatically determine the spatial hierarchy through backpropagation. The study develops an optimal deep transfer learning based on atherosclerotic plaque and calcification on intravascular ultrasound images (ODTL-APCIUI) technique. The presented ODTL-APCIUI technique aims to classify atherosclerotic plaque and calcification. To accomplish this, the presented ODTL-APCIUI technique preprocesses the IVUS images by Gaussian filtering (GF) technique. In addition, U2Net model is applied for the segmentation process with Adam optimizer based hyperparameter tuning. Moreover, the ODTL-APCIUI technique uses DenseNet-169 model for feature extraction purposes. Furthermore, the ODTL-APCIUI technique exploits stacked autoencoder (SAE) for classification process. Finally, Harris Hawks optimization (HHO) algorithm is exploited for the hyperparameter adjustment of the SAE approach. The performance assessment of ODTL-APCIUI algorithm is tested using medical images and the results are investigated under different metrics. The experimental outcomes demonstrated that the ODTL-APCIUI technique has gained better performance with maximum accuracy of 97.19%, precision of 94.09%, sensitivity of 97.17%, specificity of 97.19%, and F-score of 92.52%.