Radiomics for the Detection and Prediction of Cancer Therapy-Related Cardiotoxicity

Figure 1. Flow Diagram of Selection Strategy AUC = area under the receiver operating characteristic curve.

Figure 2. Radiomics Workflow of Model Creation This process begins with the acquisition and preprocessing of high-quality medical images, from which 3-dimensional regions of interest (ROIs) are segmented. The segmentation mask typically targets the left ventricular myocardium (LVM) rather than the cavity, as the myocardial tissue is the primary target of cardiotoxicity-related changes. Numerous quantitative features describing various properties of the ROI are extracted. Robust features are then used to build and validate a predictive/detective model, which can be integrated with clinical data for enhanced performance.23

Central Illustration. Radiomics: A Quantitative Imaging Biomarker for Cardio-Oncology The radiomics pipeline transforms routine cardiac imaging into high-dimensional quantitative signatures. Across 12 published studies in echocardiography, CT and PET/CT, radiomic models have achieved internal AUCs of 0.77 to 0.97 for the detection or prediction of cancer therapy-related cardiotoxicity. Clinical translation now requires prospective multicenter validation, broader cancer populations beyond breast, dedicated CMR investigation, and adherence to updated reporting standards (RQS 2.0, CLEAR). CMR = cardiac magnetic resonance imaging; CT = computed tomography; CTRCD = cancer therapy-related cardiac dysfunction; ECHO = echocardiography; GLS = global longitudinal strain; HF = heart failure; LASSO = least absolute shrinkage and selection operator; LVEF = left ventricular ejection fraction; ML = machine learning; PET = positron emission tomography; PPV = positive predictive value.

Figure 3. Clinical Significance of Radiomic Analysis in Cardiac Imaging (A) Ultrasomic analysis achieved from myocardial segmentation, texture feature extraction, and model output. This can aid detection of cardiomyocyte disorganization before cardiac decline. (B) CT scans often undergo deep-learning auto segmentation to extract radiomic features, which, when combined with HU texture, can predict late-onset heart failure. (C) PET/CT radiomics are extracted from myocardial SUV data and whole-heart metabolics, with cardiac homogeneity showing clinical significance (PET/CT, 95% accuracy).73,85 CT = computed tomography; FDG = Fluorodeoxyglucose/18F-FDG; LVEF = left ventricular ejection fraction; GL = gray Level; GLCM = gray level co-occurrence matrix; GLRLM = gray level run length matrix; GLS = global longitudinal strain; GLSZM = gray level size zone matrix; HU = Hounsfield units; LVM = left ventricular myocardium; PET = positron emission tomography; ROI = region of interest; SUV = standardized uptake value.