MD Automatic epicardial fat segmentation and volume quantification on non-contrast Cardiac Computed Tomography

• Autor(a) principal: Rebelo, Ana Filipa
• Data de Publicação: 2022
• Outros Autores: Ferreira, António Miguel, Fonseca, José M.
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/150293
• topic: Epicardial fat volume; Cardiac adipose tissue quantification; Automatic segmentation; Non-contrast cardiac CT; Basic image analysis; Computer-assisted tool

Epicardial Fat Volume (EFV) represents a valuable predictor of cardio- and cerebrovascular events. However, the manual procedures for EFV calculation, diffused in clinical practice, are highly time-consuming for technicians or physicians and often involve significant intra- or inter-observer variances. To reduce the processing time and improve results repeatability, we propose a computer-assisted tool that automatically performs epicardial fat segmentation and volume quantification on non-contrast cardiac Computed Tomography (CT). The proposed algorithm prioritizes the use of basic image techniques, promoting lower computational complexity. The heart region is selected using Otsu’s Method, Template Matching and Connected Component Analysis. Then, to refine the pericardium delineation, convex hull is applied. Lastly, epicardial fat is segmented by thresholding. In addition to the algorithm, an intuitive software (HARTA) was developed for clinical use, allowing human intervention for adjustments. A set of 878 non-contrast cardiac CT images was used to validate the method. Using HARTA, the average time to segment the epicardial fat on a CT was 15.5 ± 2.42 s, while manually 10 to 26 min were required. Epicardial fat segmentation was evaluated obtaining an accuracy of 98.83% and a Dice Similarity Coefficient of 0.7730. EFV automatic quantification presents Pearson and Spearman correlation coefficients of 0.9366 and 0.8773, respectively. The proposed tool presents potential to be used in clinical contexts, assisting cardiologists to achieve faster and accurate EFV, leading towards personalized diagnosis and therapy. The human intervention component can also improve the automatic results and insure the credibility of this diagnostic support system. The software hereby presented is available for public access at GitHub.