Artificial Intelligence Predicts The Efficacy Of Neoadjuvant Therapy For Rectal Cancer

Objective: In order to find the establishment of an artificial intelligence model for predicting whether rectal cancer patients benefit from neoadjuvant therapy based on the clinicopathological characteristics of rectal cancer patients and the radiohistologic(MRI)signal tags.Methods: The study was conducted on patients admitted to Sichuan Provincial People`s Hospital from January 2017 to December 2022 for rectal cancer who received preoperative neoadjuvant treatment.Collect data,establish a database,and use the Deep Learning(DL)method to build an intelligent model for predicting therapeutic effects.Use different preprocessing methods and perform feature filtering for data of different modes.Design and implement three different feature extraction networks for image information,statistical information,and natural language information.Input the extracted features of each mode into the fusion module to complement and gain modal information,Training to obtain models that can predict the efficacy of neoadjuvant therapy for rectal cancer.Based on the model,the significance method is used to analyze the interpretability of the model,synthesize various methods to obtain important feature points in the new adjuvant treatment data,and generate visual results for the importance of each feature point.The final pathological TRG(Tumor Regression Grade)obtained by the patient after receiving surgery after neoadjuvant treatment is used as the ultimate therapeutic criterion for the patient in the real world.Compare the results of AI prediction with the postoperative pathological results of patients in the real world to verify the prediction effect.Results: After collecting a total of 114 standard data in the early stage,the introduction system was randomly divided into 5 groups and cross-validation.The results show that the prediction accuracy of the model is 82.5%,and the AUC is 0.88.The diagnostic effect of the two was statistically compared by paired chi-square test(P=0.5>0.05),and the difference was not statistically significant.Through the comparison of kappa consistency test,the kappa value is 0.872>0.75,indicating that the predicted results of this model are highly consistent with the actual postoperative pathological results of patients.Results:An artificial intelligence model was established to predict whether rectal cancer patients would benefit from neoadjuvant therapy based on the clinicopathological characteristics of rectal cancer patients,radiohistologic(MRI)signal labels,and gene phenotypic status.A total of 114 standard data were input,and then randomly divided into 5 groups into efficacy prediction models and cross validated.Using information science methods such as additive feature attribution method,boosting method,and comprehensive gradient method,it is concluded that the prediction accuracy of the efficacy prediction model is 82.5%.Conclusion:It is feasible to establish an artificial intelligence model for predicting the efficacy of neoadjuvant therapy for rectal cancer using relevant data such as clinicopathological characteristics of patients with rectal cancer,radiohistology(MRI)signal tags,and to predict the efficacy of neoadjuvant therapy for rectal cancer.