Multiparametric Analysis in Knee MRI for an Early Detection of Osteoarthritis Biomarkers

Osteoarthritis and chondromalacia are diseases that are difficult to diagnose in early stages based on anamnesis, medical history, physical examination and imaging studies that are insensitive to early changes in the diseases. Chondromalacia is considered one of the risk factors for osteoarthritis which, if not treated in early stages, leads to secondary osteoarthritis. Therefore, a new diagnostic support tool is proposed based on cartigrams and the extraction of magnetic resonance image textures for subsequent analysis using radiomics. For this purpose, 101 patients were used (65 with chondromalacia patella and 36 controls) and the patellar cartilage was analyzed individually and, subsequently, the femorotibial cartilage combined with the patellar cartilage. A total of 27 patellar cartilage cartigram features and 43 femorotibial cartilage textures were obtained and analyzed using 6 predictive models: Linear SVM, Quadratic SVM, Cubic SVM, Gaussian SVM, KNN and Bagged Tree. The best result was obtained for the deep layer of the patellar cartilage by performing the cartigrams with an average AUC of 75% using the Linear SVM model and 3 features. On the other hand, for the combination of features, the best result was obtained by combining 128 grey levels of T1 with the middle layer using the Quadratic SVM model and 41 features.Clinical Relevance- This study establishes a methodology for the early diagnosis of knee osteoarthritis from chondromalacia patella based on quantitative biomarkers obtained from cartigrams and texture analysis of magnetic resonance images.