摘要:
Introduction. Knee osteoarthritis (KOA) is a highly prevalent whole joint disease that arises from various knee tissues. The infrapatellar fat pad (IPFP) is a source of knee pain in patients suffering from KOA or its precursor, patellofemoral pain (PFP). KOA diagnosis is supported by magnetic imaging resonance (MRI), from which several features can be extracted and correlated with bio- logical markers. However, few studies refer radiomics analyses for KOA (or PFP) assessment. Objectives. This study aims to automatically extract radiomic features from the femur, tibia, patella and IPFP, and assess their ability to predict KOA development over a 48-month period as well as distinguish between patients with PFP and healthy controls. Methods. Sagittal three-dimensional (3D) fast-spoiled gradient-echo (SPGR) MRI sequences with and without fat saturation (FS) from the TripleP study (n=64 PFP patients, n=70 controls) were used to study the IPFP and assess PFP. Knee structure analysis for KOA prediction was per- formed using the 3D double echo steady state water excitation MRI sequence, from the POMA dataset (n=355 KOA patients, n=355 controls). Automatic structure segmentation was achieved using nnU-Net (3D U-Net configuration), trained with 30 manually segmented knees and 5-fold cross-validation. The Workflow for Optimal Radiomics Classification toolbox was used for fea- ture extraction (n=452), and different classifiers were evaluated, namely Support Vector Machines (SVM) and Elastic Net. To compute it, SGDClassifier and Logistic Regression were selected, including hyperparameter optimization using 10-fold stratified cross-validation with a grid search and 10 repetitions. Three main models were evaluated based on: 1) patients' information (age, sex and BMI), 2) radiomic features, and 3) combination of models 1 and 2. Models' performance was evaluated using the area under the receiver operating characteristic (ROC AUC) and precision- recall (PR AUC) curves, and 95% confide