Automatic aetiology identification of left
ventricular hypertrophy from magnetic
resonance cine sequences

Clua i Sánchez, Estel

Automatic aetiology identification of left ventricular hypertrophy from magnetic resonance cine sequences

License

Citation

Clua i Sánchez, E. Automatic aetiology identification of left ventricular hypertrophy from magnetic resonance cine sequences. 2023. handle: http://hdl.handle.net/10230/57943

This citation was generated automatically.

Abstract

Left ventricular hypertrophy (LVH) consists of the thickening of the myocardium, significantly increasing the risk of cardiovascular events. LVH can be attributed to various etiologies, each with distinct prognoses and treatments. Cardiac magnetic resonance (CMR) is the principal non-invasive tool for diagnosing LVH, being cine sequences the gold standard for left ventricle characterization. However, additional time-consuming sequences are required for pathology diagnosis. Artificial intelligence (AI) has demonstrated potential in CMR analysis, creating a place for the development of AI-based automatic tools to assess the underlying aetiology of LVH using cine CMR images. This thesis aims to develop an AI-based algorithm to classify the main etiologies of LVH using cine CMR images and clinical data using a clinical database (N=304). Given the imbalanced distribution of data, two versions of the model were implemented: multi-class and binary classification. Simultaneously, the integration of automated segmentation within the classification algorithm was evaluated, utilizing a pre-trained model. Binary and multi-class classifications yielded accuracies of 69% and 62%, respectively. Segmentation results (Dice coefficient: 76.58%) underscore the necessity of a well-curated dataset that includes ground truth segmentations. Saliency maps revealed that misclassifications were primarily attributed to the algorithm’s inability to focus on myocardium. The proposed workflow represents a step forward in AI integration within clinical workflows. It is an end-to-end model ready for deployment, enabling an automated classification of LVH aetiologies in the point-of-care. Overall, this technology has been introduced to Hospital de Sant Pau being a significant advancement in enhancing diagnostic efficiency reducing costs and time.