부산대학교 도서관

Patients with myocardial infarction are at elevated risk of sudden cardiac death, and scar tissue arising from infarction is known to play a role. The accurate identification of scars therefore is crucial for risk assessment, quantification and guiding interventions. Typically, core scars and grey peripheral zones are identified by radiologists and clinicians based on cardiac late gadolinium enhancement magnetic resonance images (LGE-MRI). Manual segmentation of scar from LGE-MRI varies in size, shape, heterogeneity, artifacts, and image resolution. The process is time consuming, and influenced by the observer's experience (bias effect). Here we propose a framework that delivers an automatic segmentation pipeline to develop 3D anatomical models of the left ventricle with border zone and core scar areas that are free from bias effect. We deliver the SOCRATIS framework for automatic segmentation of cardiac left ventricle and total infarct scars. SOCRATIS involves two main pipelines; i) deep learning segmentation of cardiac regions and total infarct scar (DL-SOCRATIS) and ii) multi-atlas segmentation of cardiac regions and total infarct scar (MA-SOCRATIS). In this PhD thesis we discovered that MA-SOCRATIS was a superior technique providing high accuracy of the segmentation. However, DL-SOCRATIS was faster (less than 10 min per patient) and it does not require expert supervision, contrary to the MA-SOCRATIS. Since both pipelines are included in the SOCRATIS framework, it is possible to configure the segmentation so that duration of the analysis is traded off against high accuracy of the segmentation. SOCRATIS utilizes LGE-MR images, without the need of further multi-modal supervision (bSSFP, T2, LGE images) or training/tuning in other cohorts, contrary to the majority of the state of the art techniques. About the scar segmentation we utilized a border zone and core scar segmentation of left ventricle and total infarct scar (BZ-SOCRATIS). We tested two different segmentation AI approaches; i) an automatic and ii) semi-automatic, to segment the core scar and border zone of unhealthy tissue from LGE-MRI of the left ventricle. Our framework delivers a fully automatic segmentation pipeline to extract 3D specific patient anatomical models of left ventricle, border zone and core scar region, which can be used by clinicians to determine treatment and evaluation of the catheter radiofrequency ablation procedure.