Prediction Of High-Risk Cytogenetic Status For Multiple Myeloma Based On MRI Radiomics

Objective:This study aims to establish a prediction model or models based on MRI radiomics to evaluate the high-risk cytogenetic status of patients with multiple myeloma before treatment to stratify the risk of cytogenetic abnormalities（HRC or non-HRC）in MM patients and provide a reference for clinical treatment decisions.Materials and methods:A total of 195 patients（84 and 111 HRC and non-HRC,respectively）with MM from two centers（Lanzhou University Second Hospital and China-Japan Union Hospital of Jilin University）identified by clinical,fluorescence in situ hybridization（FISH）and undergoing preoperative magnetic resonance imaging（MRI）were retrospectively recruited.Patients with MM（71 and 88 HRC and non-HRC,respectively）were randomly divided into training（n=111）and validation（n=48）cohorts from the institution Ⅰ,while patients from institution Ⅱ（n=36）served as the external test cohort（13 and 23 HRC and non-HRC,respectively）.The regions of interest（ROIs）were assessed by T1-weighted imaging（T1WI）,T2-weighted imaging（T2WI）,and fat-suppressed T2-weighted imaging（FS-T2WI）sequences of whole spine MRI images in the sagittal position.The ROIs were determined and manually labeled by three radiologists.Features were selected for filtering using feature selection variance thresholds,Student’s t-test,redundancy analysis,Least absolute shrinkage and selection operator（LASSO）,and forward and backward stepwise regression method.Significant clinical factors were screened using univariate analysis.The radiolomics models or combined models based on T1WI,T2WI,and FS-T2WI images and clinical factors were constructed by using the logistic regression（LR）method and 10-fold cross-validation on the training cohort and firstly validated internally on the validation cohort.Nomogram performance was evaluated and compared using C-index,C-index correction by bootstrapping method,accuracy（Acc）,sensitivity（Sen）,specificity（Spc）,positive predictive value（PPV）,negative predictive value（NPV）,and Akaike information criterion（AIC）.The receiver operating characteristic curves（ROC）were used to select the most efficient radiolomics predictive model or models.And then optimal radiomics model or combined models were evaluated and compared using calibration and decision curve analysis（DCA）for the performance and clinical validity of the nomograms,respectively.Finally,the performance of the optimal model or models is tested using data from an external test cohort.In addition,Kaplan-Meier survival analysis and log＿rank tests were used to assess the prognostic value of the radiomic nomograms.Results:The efficacy of different radiomics models or combined models in predicting HRC status in MM patients before the treatment was differential.We established 14 models based on T1WI,T2WI,FS-T2WI and age(radiomics models:T₁,T₂,FT₂,T₁₊₂,FT₂₊₁,FT₂₊₂,and FT₂₊₂₊₁models;combined radiomics models:T₁+age,T₂+age,FT₂+age,T₁₊₂+age,FT₂₊₁+age,FT₂₊₂+age,and FT₂₊₂₊₁+age models).Among them,FT₂+age,FT₂₊₁+age and FT₂₊₂₊₁+age combined models were outstanding in differentiating the HRC status of MM patients in single-,double-and multi-sequence MRI images,respectively.The C-index of the training cohort and validation cohort corrected by the 1000 bootstrap method was 0.79[95%CI,0.71-0.88]and 0.80[95%CI,0.67-0.93],0.83[95%CI,0.76-0.91]and 0.84[95%CI,0.73-0.94]and0.88[95%CI,0.81-0.93]and 0.84[95%CI,0.73-0.95],respectively.The Hosmer-Lemeshow test of the calibration curves of the three models was P>0.05（P values of the training cohort were 0.9846,0.2434,0.5414,respectively;the P values of the validation cohort were 0.1979,0.1321,0.5766）.The clinical decision curve analysis（DCA）showed that FT₂+age single sequence combined model,FT₂₊₁+age double sequence combined model,and FT₂₊₂₊₁+age multi-sequences combined model could obtain good clinical net income value.Finally,in the external test cohort,the C-index of the radiomics nomogram of the FT₂+age,FT₂₊₁+age and FT₂₊₂₊₁+age combined models were 0.70[95%CI,0.53-0.87],0.76[95%CI,0.59-0.94]and 0.77[95%CI,0.61-0.92],respectively.Kaplan-Meier survival analysis showed similar prognostic values for the radiomic nomogram of the FT₂,FT₂₊₁and FT₂₊₂₊₁radiomics models and MM cytogenetic status（Log-rank test,both P<0.05,P=0.0003,P<0.0001,P=0.00011,P=0.012,respectively）.Conclusion:MRI radiomics can be used to predict HRC status in MM patients.FT₂+age,FT₂₊₁+age and FT₂₊₂₊₁+age combined models were outstanding in differentiating the HRC status of MM patients in single-,double-and triple-sequence MRI images,respectively.In addition,the age,cytogenetic status and radiomics model Rad＿score of MM patients are important factors affecting the prognosis of MM patients,which will be helpful for clinical decision-making and prognosis evaluation before treatment.