The Clinical Value Of Kaiser Score Combined With MR Radiomic Features In Differential Diagnosis Of Breast BI-RADS 4 Sub-Centimeter Masses

Objective To Explore the clinical practicality of Kaiser score and individual ADC value in diagnosing breast BI-RADS 4 sub-centimeter masses,and evaluate the auxiliary diagnostic value of different ADC values on the score.To Compare the performance difference between Kaiser score and MR Radiomic model in diagnosing breast BI-RADS 4 sub-centimeter masses,and explore the complementarity between scoring related semantic features and Radiomic features.Methods Retrospective collection of pathologically confirmed MR BI-RADS 4 sub-centimeter masses,blinded evaluation by three breast diagnostic physicians of different years based on the scoring flowchart.ADC values and TIC curves were measured separately by one physician,and the diagnostic accuracy of Kaiser and individual ADC values was analyzed through subject operating characteristic（ROC）curves.Compared different ADC values with the ADC threshold in the score to evaluate the impact of different ADC values on the score,Delong test was used to compare the differences between AUCs.Constructed a semantic feature model,Radiomic model,and a semantic feature parameter Radiomic joint model,and used the Delong test to compare the AUC differences between each model and Kaiser scores.Results A total of 173 cases were included in the study,including 67 cases of benign mass and 106 cases of malignant mass.The average diagnostic sensitivity,specificity,and accuracy of Kaiser score were 91.5%（90.6%～92.5%）,65.7%（62.7%-68.7%）,81.5%（80.9%～82.7%）,and AUC was 0.817（0.801～0.830）with a cutoff score of>4.The diagnostic sensitivity,specificity,and accuracy of individual ADC values were 55.2%and 65.1%,respectively.When 0.807 was used as the threshold,the optimal AUC was 0.604.Comparing different ADC values,ADC threshold in the original score(1.4×10^-3mm²/s)was relatively better.The Radiomic model obtained the best diagnostic performance based on Bayesian algorithm,with sensitivity,specificity,and accuracy of 86.5%,70.8%,and 80.3%in the training set,AUC of 0.839（0.762-0.914）,and sensitivity,specificity,and accuracy of 96.8%,55.0%,and 80.4%in the validation set,AUC of 0.816（0.667-0.938）.There was no statistical difference between the Radiomic model and the Kaiser score of three physicians in diagnosing AUC.The semantic feature parameters and radiomics features were complementary and could improve the diagnostic performance of the joint model.The sensitivity,specificity,and accuracy of the joint model in the training set were 89.2%,77.1%,84.4%,and AUC was 0.877（0.802-0.937）.The sensitivity,specificity,and accuracy of the validation set were 96.8%,65.0%,86.5%,and AUC was 0.863（0.732-0.962）,The Delong test showed that the joint model outperformed the semantic feature model,Radiomic model,and Kaiser score,with statistically significant differences.Conclusion Kaiser score can further differentiate breast BI-RADS 4 sub-centimeter masses,but accurate evaluation of MR imaging features and measurement of parameters are necessary.The ADC value of breast sub centimeter masses is underestimated,and comparative studies suggest that the Kaiser score’s original ADC threshold can be used for auxiliary diagnosis of breast sub centimeter masses.There is no difference in the performance of Kaiser score and MR Radiomic model in distinguishing BI-RADS 4 sub-centimeter masses.There is complementarity between score related semantic features and Radiomic features,and the combination of semantic feature parameters and Radiomic model can improve diagnostic performance.