Abstract 17987: The Added Value of Radiomics Texture Features on Late Gadolinium Enhancement Cardiac Magnetic Resonance to Detect Myocardial Inflammation in Patients With Cardiac Sarcoidosis

Abstract

Introduction: Radiomics analysis provides quantitative pixel-signal-intensity information on cardiac magnetic resonance (CMR). Hypothesis: Can radiomics improve performance of CMR for detection of myocardial inflammation in cardiac sarcoidosis (CS)? We aim to assess the added value of CMR radiomics to standard CMR image analysis for the diagnosis of active CS. Methods: We retrospectively analyzed CS patients referred for both CMR and F18-FDG-PET/CT(PET) within 90 days of each other. Myocardium was segmented on a dedicated semi-automated workflow by MIM encore software and features extracted by pyradiomics. PET were reviewed for the presence of myocardial FDG uptake, defining segments with inflammation. CMR-segments were reviewed for the presence of late gadolinium enhancement(LGE) and increased T2 signal by 2 experienced radiologists. Analysis was performed to assess for differences in radiomics features in segments with/without inflammation. ROC analysis was conducted to define the added value of radiomics for diagnosis of myocardial inflammation. Results: In 80 patients with 1360 AHA-CMR segments (mean age 58 years, 62.5% male) with CS, 276(20.3%) segments had FDG uptake on PET/CT, 252(18.5%) segments showed LGE and 54(3.9%) segments had increased signal on T2W images. Significant quantitative differences for LGE-derived radiomics features were highest for Entropy and Gray Level Cooccurrence Matrix Autocorrelation between segments with and without inflammation on PET (Table1). The presence of LGE was the best independent predictor for inflammation on PET. T2W abnormality alone was not an independent predictor from the presence of LGE. Significant improvement was seen in area under the curve when radiomics features were combined with standard CMR predictors (LGE/T2W) vs standard CMR predictors alone. Conclusions: Quantitative radiomics features on LGE-CMR have the potential to improve the ability of CMR for assessment of inflammation in CS.