Encoder-decoder architecture for ultrasound IMC segmentation and CIMT prediction

Abstract

Cardiovascular diseases (CVDs) have shown a huge impact on the number of deaths in the world. Thus, Common Carotid Artery (CCA) segmentation and Intima-Media Thickness (IMT) measurement have been significantly implemented to perform early diagnosis of CVDs by analyzing the IMT feature. In this research, we aim to implement the convolutional autoencoder model to apply semantic segmentation for Intima-Media Complex (IMC) and calculate the cIMT measurement. The results were evaluated using F1 score, precision, recall, Sorenson Dice Coefficient, and Jaccard Index. We trained the encoder-decoder architecture using 80% of the dataset and 20% was left for testing. We were able to produce results of 79.92%, 74.23%, and 60.24% for the F1 Measure, Dice coefficient, and Jaccard Index, respectively. We also calculated the IMT thickness, which was 0.54mm. Our method showed that it is robust and fully automated compared to the state-of-the-art work