Prediction of the Activity of Early Ankylosing Spondylitis using Radiomics Texture Analysis on STIR

Abstract

Abstract Purpose To evaluate the early activity of ankylosing spondylitis(AS)is very important for clinical decision. Owing to lacking of uniform standard and some other limitations, the methods for it are still difficult to practice by clinicians. The study aimed to explore the value of texture analysis of radiomics based on STIR to evaluate the activity of bone marrow edema of sacroiliac joint in early AS. Methods A total of 43 patients with early AS were collected in our hospital, including 30 males and 13 females. According to the ratio of 7:3, all the patients were randomly divided into the training cohort (n = 116) and verification cohort (n = 56). Optimal feature subsets were obtained by the Mann-Whitney U test, the minimum-Redundancy Maximum-Relevancy (mRMR), and then the least absolute shrinkage and selection operator (LASSO) using these texture feature parameters. The optimal feature subset was used to construct the final prediction model, and Radscore was obtained by summing the selected features after being weighted by their coefficients. The effectiveness of Radscore value, SPARCC score and ADC value was evaluated by ROC curve. The correlations between Radscore values and the values of SPARCC score, ADC, BASDAI, ESR, CRP, ASDAS-ESR, ASDAS-CRP were analyzed by the Spearman correlation test. Results The eight texture feature subsets were selected to obtain Redscore. In the training cohort, to differentiate early AS sacroiliac joint bone marrow edema between between active and stable group,the AUC values of Radscore, SPARCC score and ADC were 0.81,0.91,0.78, respectively. In the validation cohort, the AUC values of them were 0.87,0.89,0.85. In the two cohorts, there were no significant differences in AUC values between values of Radscore and SPARCC score, ADC (P > 0.05). There was a significant difference in AUC values betweenvalues of SPARCC score and ADC in the training cohort (P < 0.05), without statistical significance in the validation cohort (P > 0.05). The correlations were all low between Radscore values and the values of the ESR, CRP, BASDAI, ASDAS-ESR, ASDAS-CRP (P < 0.05). Conclusion Radiomics analysis based on STIR texture analysis has a good prediction for the evaluation of bone marrow edema activity of sacroiliac joint in AS.It can be a new noninvasive and objective evaluation method for AS activity, especially for the invisible lesions.