Synthetic extracellular volume from myocardial radiomics to identify transthyretin cardiac amyloidosis from routinely acquired CT angiography scans

Background Transthyretin cardiac amyloidosis cardiomyopathy (ATTR-CM) remains a significantly underdiagnosed condition, often leading to delayed treatment and management challenges. We explored the value of myocardial radiomics by computed tomography (CT) for detection of transthyretin amyloidosis cardiomyopathy. Methods The study involved 589 patients with aortic stenosis, accompanied by detailed standard pre procedural CT datasets. Radiomics were extracted from the segmented left ventricle (LV) myocardium. In Arm 1 (n=400), the method was optimized, and redundant features were removed, enhancing the accuracy of our technique. In Arm 2 (n=30), specific radiomics associated with extracellular volume (ECVCT) were identified using CT. In Arm 3 (n=159), relevant radiomics from Arm 2 were compared in patients with and without a positive bone scintigraphy scan (training cohort, n=84; validation cohort, n=75), leading to the development of a robust radiomic signature for ATTR-CM. Results In Arm 1, unsupervised clustering of patients based on radiomics was associated with significant differences in patients’ clinical profile among clusters. This clustering highlights the potential of radiomics to segregate patients based on subtle myocardial differences. In Arm 2, top radiomics correlated with ECV (panel a), were entered in a machine learning algorithm (panel b) to construct a synthetic-radiomic ECV (correlation with ECVCT: rho=0.78, p=1.2x10-6, panel c) with excellent diagnostic accuracy for high ECVCT (AUC=0.925, 95%CI: 0.825-1.000, P=0.0002, panel d). In Arm 3, a radiomic score (AmyloidRS) had good performance for ATTR-CM detection in the training (c-index 0.88, 95%CI:0.80-0.95, panels e-f) and validation cohort (c-index 0.84, 95%CI:0.69-0.98, panels g-h). The combination of AmyloidRS with clinical features maximized diagnostic accuracy for ATTR (kappa: 0.894, balanced accuracy 0.984), significantly enhancing the overall diagnostic framework. Conclusions We present a comprehensive method to calculate ECVCT from myocardial radiomics for myocardial tissue characterization. AmyloidRS, a radiomic signature for ATTR-CM, could be useful for screening for ATTR-CM from routine CT angiography scans.Figure 1