A robust radiomic-based machine learning approach to detect cardiac amyloidosis using cardiac computed tomography

Area of Science:

• Cardiology
• Radiology
• Medical Imaging

Background:

• Cardiac amyloidosis (CA) and aortic stenosis (AS) present similar clinical and imaging features, complicating diagnosis.
• CA often coexists with severe AS and carries a worse prognosis, necessitating accurate differentiation.
• Current diagnostic methods require standardization to effectively manage these overlapping conditions.

Purpose of the Study:

• To develop a robust radiomics-based pipeline for differentiating CA from AS.
• To establish a reliable, non-invasive tool for diagnosing these two cardiac pathologies.

Main Methods:

• Analysis of 107 radiomics features from cardiac computed tomography (CCT) in 30 patients (15 CA, 15 AS).
• Feature selection using various methods including LASSO and PCA, with robustness evaluated by ICC.
• Classification models (e.g., SVM) were trained and validated using leave-one-out cross-validation and data augmentation.

Main Results:

• Ninety-two robust radiomic features were identified.
• A support vector machine classifier achieved 0.93 accuracy, sensitivity, and specificity using PCA and a 0.95 correlation threshold.
• Principal components were primarily driven by textural, first-order, and shape/size features.

Conclusions:

• Radiomics analysis of CCT images shows potential as a non-invasive tool to differentiate CA from AS.
• This approach can aid in the early and accurate diagnosis of these conditions using routine imaging.