Magnetic Resonance Big Data Quantitative Study Of Knee Osteoarthritis

PURPOSE: Osteoarthritis is a heterogeneous disease with multiple causes,clinical presentations and pathological manifestations.The purpose of this study is to use the deep learning algorithm to achieve fully automatic classification and segmentation of knee joint cartilage through a large number of knee joint magnetic resonance images of healthy people and osteoarthritis patients,evaluating its accuracy,and to establish a knee joint big data artificial intelligence diagnosis software.By means of the data processing method,calculate the cartilage thickness of normal Chinese people at different ages based on the accurate segmentation of the knee joint magnetic resonance image.Furthermore,based on big data analysis,establish the knee cartilage thickness curve of healthy Chinese people at different ages(16-65 years old).METHODS: This study adopted a retrospective analysis of a total of 590 data to establish an automatic segmentation and classification system for knee cartilage.Among them,557 were used for automatic classification training,and 33 were used for automatic segmentation(27training data,6 test data),using the deep learning method and the neural network model--V-type network and Inception network for automatic segmentation and classification.The segmentation results were evaluated by the Dice similarity coefficient,and the classification results were verified by a five-fold cross-validation,and integrate these processes with deep convolutional neural networks,single-image super-resolution algorithms,and very deep network super-resolution algorithms.In addition,2365 healthy human(16-65 years old)knee joint magnetic resonance data were obtained.The average thickness of cartilage at all ages and femoral surfaces was obtained by image segmentation,thickness calculation,registration and regression.RESULTS: The DSC values of the automatic segmentation of the femur,tibia and fibula of the knee were respectively 0.9856±0.0053,0.9806±0.0037 and 0.9121±0.0251,higher than 0.9,and higher than manual labeling.The sensitivity,specificity and accuracy of the five-fold data are higher than 0.93,and all of the AUC values of the five-fold data are higher than 0.98 and close to 1.At the optimal threshold of 0.0014,the total sensitivity and specificity were 0.9462 and 0.9527,respectively.The total accuracy was 0.9521,and the AUC value was 0.9859.The knee joint big data artificial intelligence diagnosis software can automatically segment the image within 5s,and the automatic classification speed of cartilage lesions is within 1s.The thickness of the femoral cartilage canbe calculated in routine clinical magnetic resonance sequences within 5minutes.CONCLUSIONS: This study confirms the feasibility of establishing a knee joint big data artificial intelligence diagnosis system combined with deep learning algorithm,and using the big data analysis we establish the knee femur cartilage curve of healthy Chinese people at different ages(16-65 years old),providing early supports for the important role that artificial intelligence will play in supporting physicians and healthcare systems in the future.