Automatic Segmentation of the Left Ventricular Myocardium in Cardiac CT Scans using AI

Heart failure (HF) is a leading cause of morbidity and mortality. Next to coronary artery disease, inherent (genetic or acquired) disorders of the heart muscle are accountable for almost half of the cases of HF. Easily deployable technologies that could be used to recognize subclinical heart muscle disease before the clinical development of HF would be valuable as population screening tools for early detection of myocardial disease. Artificial Intelligence (AI) and Machine Learning (ML) on computed tomography (CT) medical scans of the heart can provide a fully automated method for myocardial tissue characterization by segmenting images and, in turn, extracting radiomic features. We present a method to segment the left ventricular (LV) myocardium in an automatic manner. CT scans from healthy individuals were segmented manually and used to train and validate various AI models. The four best-performing models were selected and combined to a weighted AI segmentation model with a DICE-Score (F1-Score) of 0.907. This automated model for segmentation of LV myocardium can be used as part of an analysis pipeline for myocardial tissue characterization by imaging.