Novel Self-Calibrated Threshold-Free Probabilistic Fibrosis Signature Technique for 3D Late Gadolinium Enhancement MRI.

Abstract

Myocardial fibrosis, marked by excessive collagen buildup in the heart, is a crucial severity marker of heart muscle injury in several heart diseases, such as myocardial infarction, cardiomyopathies, and atrial fibrillation (AF). It is also vital for evaluating the efficacy of induced scarring (dense fibrosis) post-interventions, such as catheter ablation for AF. Cardiac MRI emerged as the gold standard for evaluating myocardial fibrosis and scarring for diagnosis and intervention planning. However, existing 3D cardiac MRI (CMR) fibrosis analysis methods are unreliable as they rely on variable thresholding and suffer from a lack of standardization and high sensitivity to typical MRI uncertainties. Importantly, these methods quantify severity based on fibrosis volume alone while ignoring the unique MRI characteristics of fibrosis distribution, which could better inform on disease severity. To address these limitations, we propose a novel thresholdfree and self-calibrating probabilistic method named "Fibrosis Signatures" for a comprehensive and reliable fibrosis analysis of 3D MRI cardiac images. Through a novel efficient (linear complexity) probabilistic encoding of 'multibillion' MRI intensity disparities into standardized probability density function, our method derives the patient's unique fibrosis signature profile and index (FSI). Our approach goes beyond mere measuring of fibrosis volume; it encodes both the extent and the unique MRI characteristics of fibrosis distribution beyond mere entropy for a more detailed evaluation of fibrosis burden/severity. Our self-calibrating design effectively adjusts for MRI uncertainties like noise, low spatial resolution, and segmentation errors to ensure robust and reproducible fibrosis evaluation pre- and post-intervention. Validated in numerical phantom and 143 in vivo MRI scans of AF patients and compared to five baseline methods, our method showed strong correlations with traditional volume measures of pre-intervention fibrosis and post-intervention scar and was up to 9- times more reliable and reproducible, highlighting its potential to enhance cardiac MRI's utility.