Computer Vision News - February 2020

Summary RSIP Vision's Project 18 Echocardiography plays a crucial role in the diagnosis and management of cardiovascular disease. As the only imaging modal ity with real -time imaging of the heart, echocardiography al lows immediate detection of pathologies, giving it a place across the entire cardiac patient care pathway. Several cardiac ultrasound technologies are avai lable: Doppler analysis, M-mode echocardiography, two-dimensional transthoracic echocardiography (TTE) , transesophageal echocardiography (TEE) , and 3-D echocardiography. However, imaging analysis is dependent on operator and interpreter experience. Cardiovascular imaging modal ities are being disrupted by deep learning algorithms that are improving diagnostic accuracy and cl inical management. Two Challenges in Echocardiography Echo technicians (sonographers) must identi fy an ideal pathway the ‘thoracic window’ that is relatively free from interfering obstacles, posing chal lenges to obtaining an accurate echocardiogram. This di ff iculty was i l lustrated in one recent study that found echocardiographic qual ity inadequacy in 24% of imaging studies. Additional ly, interpreting these mani fold images rel ies on human experts, making results prone to error. DL signi f icantly improves the speed and accuracy of image segmentation and classi f ication, thereby streaml ining cardiac care. 1.Technical Challenge: Left Ventricle Segmentation Accuracy LV segmentation to assess cardiac structure and function rel ies on increasingly data-heavy ultrasound imaging, which is the gold standard method, in heart examination. The accuracy of these images is affected by variabi l ity, motion, noise, and poor contrast, and the vast quantity of images from multiple views poses a chal lenge to workf low eff iciency. LV segmentation benef its from automation, propel l ing the development of state-of-the-art software solutions, however, these Deep Learning for Echocardiography