Recognition Of Tumor Cells In Patients With Early Death Of Multiple Myeloma By Artificial Intelligence Transfer Learning Purpose

Purpose Establish a classic machine learning model and a deep learning model for judging the prognosis of multiple myeloma patients based on the morphology of tumor cells,aiming to identify early death high-risk multiple myeloma patients through the morphology of tumor cells,and provide a basis for the stratified treatment of myeloma.Materials and Methods The bone marrow smears of 20 patients with early death of multiple myeloma who were admitted to the China-Japan Union Hospital of Jilin University from 2011 to 2020 and the bone marrow pictures of 20 patients with non-early death who were admitted in the same period were collected as controls,and 40 pictures were collected.The bone marrow smear is scanned for the whole image;the myeloma cells of the two groups are segmented by the bone marrow pathological diagnosis physician;the classic machine learning model and the migration learning model are established based on the morphology of the myeloma cells,and the myeloma cells of the two groups of patients are trained.According to the obtained confusion matrix and receiver characteristic curve,the accuracy of the two models in judging the prognosis of patients with multiple myeloma is evaluated.Results1.Obtained WSI data on bone marrow smears of 20 early death patients and 20non-early death patients.2.Diagnosis of bone marrow pathomorphology divided 2535 myeloma cells from patients with early death and 1608 myeloma cells from patients with non-early death;3.Construct a classic machine learning model for identifying the prognosis of multiple myeloma patients based on the morphology of myeloma cell nuclei:3.1 The accuracy of the confusion matrix was 76.9%,and the AUC value of the ROC curve was 0.87.3.2 Classical machine learning models can judge the prognosis of patients with multiple myeloma based on the morphology of the nucleus to a certain extent.4.Construct a deep learning model based on the morphology of myeloma cells to identify the prognosis of patients with multiple myeloma:4.1 Based on migration learning,a deep learning model for judging the prognosis of patients with multiple myeloma based on the morphology of tumor cells was completed.4.2 Evaluation of the deep learning model for judging the prognosis of myeloma patients based on the morphology of myeloma cells: the accuracy of the confusion matrix was 89.66%;the ROC curve AUC value was 0.868.Conclusions1.Classical machine learning based on the morphology of tumor cell nucleus to determine the prognosis of patients with multiple myeloma can determine the prognosis of patients to a certain extent.2.The transfer learning model of plasma cell morphology can be used to predict early death in myeloma patients.