Machine Learning-based MRI Radiomics Signature Of Subchondral Bone For Prediction Of Incidence And Progression Of Knee Osteoarthritis

Identification of risk factors in the early phase of knee osteoarthritis(OA)(before irreversible lesions occur)is crucial for performing preventive strategies to prevent or delay the progression of the disease.Previous studies have confirmed that subchondral bone plays a critical role in the pathophysiology of knee O A.The aim of this study was to develop and validate an MRI-based radiomics signature from subchondral bone with machine learning for predicting incidence and progression of knee OA.This study consisted of three chapters.The first chapter aimed to develop and validate an MRI-based radiomics signature from subchondral bone to distinguish between knees without and with OA.This chapter involved 200 knees(139 knees with medial tibiofemoral OA,3 knees lateral tibiofemoral OA,31 knees with bilateral tibiofemoral OA,27 normal knees),which were divided into training set(168 knees)and validation set(32 knees).The radiomics signatures were calculated with the radiomics features of subchondral bone extracted from T1W/TSE and T2-spair MRIs.The least absolute shrinkage and selection operator(LASSO)method was used to obtain the most significant features.Accuracy(ACC),sensitivity and specificity,the area under the receiver operating characteristic curve(AUC)are used for quantifying the discrimination.The calibration performance of radiomics signature,which indicates the agreement between the models and actual outcome,was evaluated with calibration curve and Hosmer-Lemeshow goodness of fit test.The results showed radiomics signature generated an AUC of 0.797(95%CI,0.689-0.905)and 0.711(95%CI,0.689-0.905)in the training and validation cohorts of medial tibiofemoral OA respectively,radiomics signature generated an AUC of 1.000(95%CI,1.000-1.000)and 0.667(95%CI,0.400-0.934)in the training and validation cohorts of lateral tibiofemoral OA respectively.Above results suggested that MRIbased radiomics signature from subchondral bone could distinguish between knees without and with O A.The second chapter aimed to develop and validate an MRI-based radiomics signature from subchondral bone for predicting incident knee OA and also developed a nomogram incorporating radiomics signatures and two clinical factors(BMI and WOMAC pian score)that could provide a visual presentation of prediction model.This chapter involved a nested case-control study design that included 347 case knees and 347 control knees.554 knees were assigned to the training set with the rest 140 knees assigned to the validation set.Discrimination and calibration were estimated by above methods and clinical usefulness of nomogram was determined with decision curve analysis(DCA).The results showed that radiomics signature generated an AUC of 0.848(95%CI,0.816-0.880)and 0.657(95%CI,0.567-0.747)in the training and validation cohorts respectively.The radiomics nomogram integrating the radiomics score and clinical factors(BMI and WOMAC pain score)demonstrated favorable discrimination with AUC of 0.858(95%CI,0.828-0.889)and 0.743(95%CI,0.6600.826)in the training and validation cohorts respectively.The calibration curve showed a good agreement.The DCA demonstrated that the radiomics nomogram had a more profitable performance than the radiomics signature in the threshold probability range of 0 to 0.81.Above results suggested that MRI-based nomogram integrated radiomics signature from subchondral bone and two clinical factors(BMI and WOMAC pian score)could be utilized to prediction of incident knee OA.The third chapter aimed develop and validate an MRI-based radiomics signature from subchondral bone for predicting total knee replacement and also developed a nomogram incorporating radiomics signatures and WOMAC score that provide a visual presentation of prediction model.This chapter involved 208 knees,including 105 case knees(received TKR in 5 years follow-up)and 103 control knees.152 knees were assigned to the training set with the rest 56 knees assigned to the validation set according to the clinical centers where they were recruited.Discrimination,calibration and clinical usefulness were estimated by above methods.The results showed that radiomics signature generated an AUC of 0.950(95%CI,0.918-0.982)and 0.706(95%CI,0.567-0.845)in the training and validation cohorts respectively.The radiomics nomogram integrating the radiomics score and WOMAC score demonstrated favorable discrimination with AUC of 0.971(95%CI,0.946-0.996)and 0.798(95%CI,0.683-0.914)in the training and validation cohorts respectively.The calibration curve showed a good agreement.The DCA demonstrated that the radiomics nomogram had a more profitable performance than the radiomics signature in the threshold probability range of 0 to 1.0.Above results suggested that MRI-based nomogram integrated radiomics signature from subchondral bone and WOMAC pian score could be utilized to prediction of TKR.In summary,this study demonstrated that the MRI-based radiomics signature from subchondral bone could be utilized for predicting incidence and progression of knee OA.