Intelligent Prediction Of The Efficacy Of Chemotherapy For Colorectal Cancer

BACKGROUNDColorectal cancer(CRC),a high incidence of malignant tumors worldwide[1,2]Chemotherapy is an important means to inhibit tumor growth after surgery,which greatly improves the survival of patients[1,2].But different individuals respond differently to chemotherapy.Only 20%of them actually receive clinical benefit,and most still have chemotherapy non-response(resistance).Regarding the evaluation of the effect of chemotherapy,the detection of tumor marker(CA199),carcinoembryonic antigen(CEA)and periodic imaging examination after the course of treatment are the main clinical methods,and to a certain extent,the functional mechanism,drug resistance genes and molecules related to chemotherapy non-response can be predicted The research on targets is too numerous to mention,but it still needs further integration and verification.It is an urgent clinical need to evaluate the efficacy of chemotherapy before chemotherapy.Therefore,there is an urgent need to add new non-invasive and efficient prediction methods to accurately screen suitable populations.Artificial Intelligence(AI),with its powerful image recognition and feature analysis technology,has brought a new perspective to innovative colorectal cancer prognosis assessment methods.More and more studies at home and abroad expect to combine AI to solve such clinical problems.At present,in the field of colorectal cancer,the use of AI platforms to assist in the assessment of prognosis is mainly based on imaging omics[5-7],and the association of clinical features and prognostic status based on pathology with AI algorithms has not been effectively carried out.However,studies have confirmed that AI-guided machine learning can be used to analyze pathological pictures such as prostate cancer to predict the risk of disease progression in patients[8].The above research will undoubtedly open up new ideas for the evaluation of chemotherapy response in colorectal cancer,that is,based on digital pathological images combined with AI algorithms,to construct accurate and non-invasive prediction models,"see" micro or sub-vision that clinical and pathologists cannot Digital pathological biomarkers for predicting application value have enriched clinical decision-making weights and achieve accurate prediction of chemotherapy response,which has huge potential application value.OBJECTIVETo explore the intelligent application model of AI-assisted colorectal cancer chemotherapy and the feasibility and accuracy of the intelligent pathological prediction model of colorectal cancer chemotherapy.METHODSA total of 89 eligible samples were collected and divided into response group/partial response group,non-response group/progressive group according to the conditions,and digital pathological images were obtained.Method 1 was by labeling tumor tissue and using ResNet50 model to extract features.Method 2 was by labeling tumor tissue And tumor microenvironment(extratumor stroma,tumor infiltrating immune cells,tumor-associated blood vessels),using Xception?InceptionV3?and ResNet50 fusion models to extract features,clustering and classification using algorithms,constructing a model for predicting the efficacy of chemotherapy Model accuracy.RESULTS1.The accuracy of the intelligent pathological prediction model of chemotherapy efficacy based on the characteristics of colorectal cancer tumor cells on the training set and validation set is about 0.99(99%)and 0.74(74%),The non-response group/progress group,and the accuracy rates of all sample groups are about 0.67(66.67%),0.90(90%),and 0.80(80.00%);the areas under the ROC curve on the training set and test set are 1.00(AUC=1.00),0.74(AUC=0.74).2.The accuracy of the intelligent pathological prediction model based on colorectal cancer tumor cells and their microenvironment(tumor interstitial,infiltrating immune cells in the tumor,tumor-associated blood vessels)on the efficacy of chemotherapy in the training set and validation set is about 0.92(92%)and 0.78(78%),the accuracy rate can reach 0.83(83%)when the test is the best;the area under the ROC curve on the training set and the test set is 1.00(AUC=1.00),0.87(AUC=0.87).CONCLUSIONSThe accuracy of the intelligent pathological prediction of colorectal cancer chemotherapy is high and the effect is good.The accuracy of the method based on the characteristics of the tumor or tumor and its microenvironment is about 75%,and in good cases it can be as high as 80%or more.Sex,the predicted results have clinical significance.The accuracy of the intelligent pathology prediction model based on the colorectal cancer tumor cells and their microenvironment(tumor interstitium,tumor infiltrating immune cells,tumor-associated blood vessels)is superior to the single tumor cell-based features.