Radiomics-based method for diagnosis of calciphylaxis in patients with chronic kidney disease using computed tomography.

Abstract

Calciphylaxis is a rare, life-threatening condition that affects patients with chronic kidney disease (CKD) undergoing dialysis. Skin biopsy as the gold standard causes ulceration, bleeding, or infection. This study aimed to develop radiomic methods using CT as a noninvasive method for calciphylaxis diagnosis. The confirmed calciphylaxis patients (Group I), pathologically confirmed non-calciphylaxis patients (Group II), and CKD patients (Group III) from October 1, 2017, to November 30, 2019, were enrolled. Training: 70% of patients of Group I and all Group III. Test: 30% of patients of Group I and all Group II. ROI was set at the skin lesion including the soft tissue. First-order and texture features were extracted from each lesion unit. CT-based radiomic models were on the basis of logistic regression (LR) and support vector machine (SVM). Additionally, model performance was evaluated in the test dataset and compared with the plain radiography and bone scintigraphy. In total, 124 lesions and 38 lesions were identified in training and test datasets. Radiomic models were effective in detecting calciphylaxis in patients with CKD, with AUCs of 0.93 (95% CI: 0.924-0.953) and 0.93 (95% CI: 0.921-0.953) (SVM and LR) in test. The SVM model manifested a sensitivity and specificity of 0.89 and 0.8, and 0.78 and 0.90, at high-sensitivity and high-specificity operating points, respectively. Similar performance was found in the LR model. Radiomic models were more effective than plain radiography and bone scintigraphy (Delong test, P<0.05). Verification studies showed the features which manifested the real variability of lesions. In this research, it primarily developed a radiomic method for noninvasive detection of calciphylaxis in patients with CKD. Through this method, calciphylaxis can be detected when invasive procedures are not feasible.