Automatic quality assessment of end-to-side anastomoses using epicardial ultrasound images recorded from coronary artery bypass graft surgeries
Translated title
Automatisk kvalitetssikring af end-to-side anastomoser ved anvendelse af epicardielle ultralydsbilleder optaget under koronararterie bypass graft operationer
Authors
Kristensen, Mads ; Rasmussen, Birgithe Kleemann ; Poulsen, Linette Helena
Term
4. term
Publication year
2019
Submitted on
2019-06-06
Pages
74
Abstract
Indledning: Intraoperativ kvalitetskontrol ved hjertekirurgi er vigtig for at sikre resultatet og undgå efterfølgende reoperation. Ved koronar bypass-operation (CABG) skaber kirurgen en omkørselsvej for blodet ved at sy en bypass-graft til en kranspulsåre; sammensyningsstedet kaldes en anastomose. CABG er forbundet med tekniske komplikationer, men er én af de få større karkirurgiske procedurer, som ikke rutinemæssigt kvalitetstjekkes under selve indgrebet. Udbredte metoder til at vurdere anastomosekvalitet er enten upraktiske eller kan give et forkert indtryk. Epikardiel ultralyd (EUS) – ultralyd direkte på hjertets overflade – kan vise anastomosens struktur og er et lovende alternativ. Tolkningen er dog ofte subjektiv, hvilket kan føre til manglende eller unødvendige revisioner. Formålet med projektet var derfor at udvikle en objektiv metode til kvalitetsvurdering. Metoder: Der var tilgængelige 367 EUS-billeder af anastomoser; 96 blev brugt til at udvikle metoden. Arbejdsgangen omfattede kar-detektion, segmentering af karlumen (hulrummet i karet) og estimering af åbenhed (patency). To segmenteringsmetoder blev afprøvet og sammenlignet: en lokal-fase-baseret “snake” og Chan–Vese. Resultater: På de resterende 271 testbilleder fandt metoden fulde eller delvise karkonturer i 89,67% af billederne. For de tilstrækkelige detektioner var den gennemsnitlige Dice-koefficient (overlap med ekspertkonturer) 0,8134 med den lokal-fase-baserede “snake” og 0,8187 med Chan–Vese. Dette muliggjorde åbenhedsvurdering i op til 135 billeder. Ved sammenligning af åbenhedsvurderingerne med tilsvarende manuelle annotationer var den højeste overensstemmelse 88,15% med Chan–Vese-segmenteringen. Konklusion: Metoden kunne automatisk finde anastomosekar på EUS-billeder og vurdere deres åbenhed. Præcis kantplacering i forhold til ekspertmarkeringer er dog fortsat vanskelig på grund af billedartefakter i EUS-billederne.
Introduction: Checking the quality of cardiac surgery during the operation is important to ensure good outcomes and avoid later reoperation. In coronary artery bypass grafting (CABG), surgeons create a new route for blood by sewing a bypass graft to a coronary artery; the join is called an anastomosis. CABG can involve technical complications, yet unlike many other major vascular procedures it is not routinely checked intraoperatively. Common methods for assessing anastomosis quality are either impractical or can be misleading. Epicardial ultrasonography (EUS)—ultrasound applied directly to the heart’s surface—can show the structure of the anastomosis and is a promising alternative. However, interpretation is often subjective, which can lead to missed or unnecessary revisions. The aim of this project was to develop an objective method for quality assessment. Methods: A total of 367 EUS frames showing anastomoses were available; 96 frames were used to develop the method. The pipeline included vessel detection, segmentation of the vessel lumen (the inner open space), and estimation of patency (how open the vessel is). Two segmentation techniques were tested and compared: a local-phase-based snake and Chan–Vese. Results: On the remaining 271 test frames, the method detected full or partial vessel structures in 89.67% of images. For frames with sufficient detections, the average Dice coefficient (overlap with expert outlines) was 0.8134 for the local-phase-based snake and 0.8187 for Chan–Vese. This enabled patency estimation in up to 135 frames. When patency estimates were compared with corresponding manual annotations, the highest agreement was 88.15% using Chan–Vese segmentation. Conclusion: The proposed method can automatically detect anastomosis vessels in EUS images and estimate their patency. Precisely matching expert-drawn edges remains challenging due to artifacts in the EUS frames.
[This abstract was generated with the help of AI]
Keywords
Documents