SPECT Sol Ventrikül Bölütlenmesi Kullanılarak Makine Öğrenmesi Yöntemleri ile Kalbin Uzun Ekseni Çıkarım

Abstract

Segmentation of myocardial tissue in SPECT (Single Photon Emission Computed Tomography) left ventricle images is a crucial problem for assisting diagnosis. Non-AI-based segmentation models in this field typically segment by first extracting key points such as the apex and base of the heart, as well as lines like the long axis, based on predefined assumptions. However, these models perform poorly in cases where SPECT images are noisy. In contrast, AI-based models, which are more robust to noise, can perform segmentation without requiring any predefined points or axes. In clinical practice, segmented heart images are examined by experts using short-axis, vertical long-axis, and horizontal long-axis views. Therefore, determining the long axis of the left ventricle is of critical importance. As a novel contribution to the literature, this study aims to extract the long axis from binary segmentation images in AI-supported SPECT left ventricle segmentation models - a missing aspect in current approaches. The deep learning model we developed determines the symmetry axis in given 3D binary segmentation images and extracts clinically important cross-sections from long and short axes for diagnostic evaluation. © 2025 Elsevier B.V., All rights reserved.