A radiomics method based on MR FS-T2WI sequence for diagnosing of autosomal dominant polycystic kidney disease progression.

OBJECTIVE We aimed to construct/validate a radiomics method based on MR FS-T2WI sequence for the evaluation of kidney function in patients with autosomal dominant polycystic kidney disease (ADPKD). PATIENTS AND METHODS The clinical data and MRI images of 114 patients with ADPKD were retrospectively analyzed. With a glomerular filtration rate of 60 mL/min per 1.73 m2 as the cutoff value, patients were divided into two groups, where there were 59 patients with GFR ≥60 mL/min per 1.73 m2 (including CKD1 and CKD2 phase) and 55 patients with GFR <60 mL/min per 1.73 m2 (including CKD3 phase and higher). All patients underwent the 3.0T MR scan of the kidney. Then, the kidney were delineated layer by layer based on the FS-T2WI sequence to obtain the volume of interest (VOI) for radiomics features extraction. The optimal radiomics features were selected by least absolute shrinkage and selection operator (LASSO). Three kinds of data modality including the pure clinical data, the pure image data and the clinical-image fused data were utilized to establish three types of models (clinical, image and with their combination) separately by five machine learning classifiers: k-nearest-neighbors (KNN), support vector machine (SVM), logistic regression (LR), random forests (RF) and multi-layer perception (MLP). Receiver operating characteristic (ROC) curve, areas under the curve (AUC), sensitivity, specificity and precision were employed to evaluate the model's effectiveness to diagnosis the glomerular filtration rate of patients with ADPKD based on different models. Besides, Delong test was applied to compare ROCs between models. RESULTS 960 radiomics features were extracted from each VOIs, and clinical information included the gender and age of each patient. After feature selection, 23 and 21 features based on pure image data and clinical-image fused data were independently used to construct models for the kidney function evaluation. The clinical-image fused model (AUC=0.89) has better performance than the pure image model (p=0.046) and pure clinical model (p<0.001). Clinical-image fused model based on LR classifier showed the best diagnostic efficiency, with AUC=0.89, sensitivity=0.8867 and specificity=0.7959. CONCLUSIONS The MR FS-T2WI radiomics analysis based on clinical-image fused model is instrumental in evaluating and predicting the kidney function of patients with polycystic kidney disease.