Application of Morphogo in the detection of minimal residual disease in multiple myeloma

Abstract:

Objective To explore the application prospects and problems of artificial intelligence bone marrow cell recognition system，Morphogo，in the detection of minimal residual disease（MRD） of multiple myeloma.Methods A total of 65 cases of MRD of multiple myeloma confirmed by multiparameter flow cytometry（MFC） were collected，and their bone marrow Wright's staining smears were obtained. All bone marrow smears were automatically scanned and classified by Morphogo based on artificial intelligence platform. The positive threshold of MRD of AI and cytomorphology in multiple myeloma was set as the proportion of plasma cells，which was greater than 4.4%. The cases were divided into I 500 group，I 1000 group and I 2000 group according to the number of AI automatically recognized cells. The results of AI-MRD，morphological-MRD（M-MRD） and MFC-MRD in each group were tested by Kappa consistency test，and the sensitivity，specificity and accuracy of each group were calculated. Taking MFC-MRD and M-MRD results as the gold standard，the receiver operating characteristic (ROC) curve of AI-MRD was drawn and its area under the curve (AUC) value was calculated.Results After grouping，with the increase of the number of recognized cells，the Kappa value，sensitivity，specificity，accuracy and AUC of AI-MRD vs. MFC-MRD and AI-MRD vs. M-MRD increased. The Kappa consistency test of AI-MRD vs. MFC-MRD in I 2000 group showed that the Kappa value was 0.500（P=0.013），sensitivity was 71%，specificity was 80%，and accuracy was 75%. The Kappa consistency test results of AI-MRD vs. M-MRD showed that the Kappa value was 0.667（P=0.001），sensitivity was 100%，specificity was 75%，and accuracy was 83%. When MFC-MRD results were taken as the diagnostic criteria，the AUC of AI-MRD in I 2000 group was 0.800（P=0.002，95%CI=0.588-0.934），and the AUC of M-MRD was 0.779（P=0.005，95%CI=0.564-0.921）.Conclusion The detection of MRD of multiple myeloma by Morphogo has the characteristics of high accuracy，high speed and low cost. In the follow-up development，it should be considered to develop technologies such as cell histochemical staining and cellular immunity，so as to improve the diagnostic accuracy of MRD of artificial intelligence multiple myeloma.