Aortic valve segmentation using deep learning

• Main Authors: Lai, Khin Wee, Shoaib, Muhammad Ali, Chuah, Joon Huang, Ahmad Nizar, Muhammad Hanif, Anis, Shazia, Ching, Serena Low Woan
• Format: Conference or Workshop Item
• Published: 2021
• Subjects: QH301 Biology; R Medicine; T Technology (General)
• Online Access: http://eprints.um.edu.my/35403/

Aortic stenosis is the most common type of valvular heart disease (VHD), requiring echocardiography examination for diagnosing and monitoring of the patient. Segmentation of the aortic valve is one of the crucial medical tasks as it helps in different cardiac treatments, e.g. in aortic valve replacement. Manual segmentation is tedious and depends upon the expertise of clinicians so automated segmentation of aortic valve is primarily significant. Deep learning is a viable solution for the automatic segmentation of the aortic valve. Unfortunately, there is lacking knowledge in the application of deep learning in echocardiography. This study proposes a deep learning technique to segment the aortic valve. Echocardiography data of 58 patients for training and neural networks evaluation were obtained from National Heart Institute (IJN). Bi-Directional ConvLSTM U-NET (BDCU-Net),and UNet were trained to segment planimetry aortic valve area. BDCU-Net had the Fl-score 91.092%, followed by UNet90.618 degrees A. The results show that BDCU-Net performance is better than U-Net.