Computed tomography-derived myocardial radiomics for detection of transthyretin amyloidosis in patients with severe aortic stenosis

We explored the value of myocardial radiomics by computed tomography angiography (CTA) for detection of transthyretin amyloidosis cardiomyopathy (ATTR-CM). The study included 589 patients with aortic stenosis and CTA datasets. Radiomics were extracted from LV myocardium. Arm 1 (<i>n</i> = 400) served for method optimisation and removal of redundant features. In Arm 2 (<i>n</i> = 30), we identified radiomics associated with extracellular volume by CT (ECV_CT); in Arm 3 (<i>n</i> = 159), radiomics were compared in patients with/without positive bone scintigraphy scan (training cohort, <i>n</i> = 84; validation cohort, <i>n</i> = 75) to build a radiomic signature for ATTR-CM. In Arm 1, unsupervised clustering of patients based on radiomics was associated with significant differences in patients’ clinical profile among clusters. In Arm 2, we constructed a radiomic-based ECV (correlation with ECV_CT: rho = .78, <i>p</i> = 1.2 x 10⁻⁶) with excellent diagnostic accuracy for high ECV_CT (AUC = .925, 95%CI: .825–1.000, <i>p</i> = .0002). In Arm 3, a radiomic score (AmyloidRS) had good performance for ATTR-CM detection in the training (c-index .88, 95%CI: .80–.95) and validation cohort (c-index .84, 95%CI: .69–.98). When combined with clinical features, AmyloidRS maximised diagnostic accuracy for ATTR (kappa: .894, balanced accuracy .984). We present a radiomic method for myocardial tissue characterisation in patients with severe aortic stenosis which enables ATTR-CM detection from standard CTA scans.