Characterization of interstitial diffuse fibrosis patterns using texture analysis of myocardial native T1 mapping

Area of Science:

• Cardiology
• Medical Imaging
• Computational Pathology

Background:

• Myocardial fibrosis patterns vary significantly between hypertrophic cardiomyopathy (HCM) and dilated cardiomyopathy (DCM).
• HCM fibrosis is typically patchy and regional, while DCM fibrosis is diffuse and global.
• Differentiating these patterns is crucial for accurate diagnosis and management.

Purpose of the Study:

• To investigate the utility of texture analysis on native T1 mapping for differentiating fibrosis patterns in HCM and DCM.
• To assess if texture features can accurately classify patients with HCM, DCM, and healthy controls.

Main Methods:

• Prospective acquisition of native myocardial T1 mapping images from 321 subjects (116 HCM, 140 DCM, 65 controls).
• Extraction of 152 texture features from T1 maps using texture descriptors.
• Selection of seven key features for pattern identification and classification using support vector machines (SVM) and regression models on training and testing datasets.

Main Results:

• Texture analysis achieved 85.2% accuracy in classifying subjects in the training set (AUC=0.93 for all groups), significantly outperforming pattern-independent global T1 mapping (51.1% accuracy, P<0.001).
• The model demonstrated excellent reproducibility and generalizability, achieving 89.3% accuracy on the testing dataset.
• Pattern-derived features showed excellent intra- and inter-observer reliability.

Conclusions:

• Texture analysis of native T1 mapping effectively characterizes myocardial fibrosis patterns in HCM and DCM.
• This approach provides valuable diagnostic information beyond average native T1 values.
• Texture analysis holds promise for improving the non-invasive differentiation of cardiomyopathies based on fibrosis patterns.