Prediction Of Cancer Metastasis In Colon Cancer Patients Based On Enhanced CT Images

Colon cancer is a common malignancy,and early detection and treatment can lead to a longer survival for most patients.Otherwise,the risk of tumor metastasis will greatly affect the survival and prognosis of patients.Therefore,it is important to evaluate the risk of cancer cell metastasis in colon cancer patients and to take effective personalized treatment to improve their prognosis.It is of great significance to use medical imaging data to predict cancer metastasis in colon cancer patients.In this paper,with the focus on whether cancer metastasis occurs in patients with colon cancer,the correlation between CT value characteristics of cancer in patients with colon cancer and whether cancer metastasis occurs is studied by cluster analysis method and machine learning classification algorithm.In this study,173 patients with colon cancer were treated with enhanced CT images.We extracted the gray matrix of the lesion from the arterial phase and portal pulse phase of the CT sequence map to obtain the two-dimensional CT value characteristics of each pixel in the patient’s tumor.On this basis,firstly,the K-means clustering algorithm is applied to extract the features of the overall patients;secondly,the consensus clustering algorithm is used to determine the number of tumor clustering categories of individual patients;thirdly,traditional clustering methods are used,such as: Hierarchical clustering,K-means clustering and fuzzy C-means clustering,as well as the deep clustering algorithm based on variational autoencoder in the deep clustering algorithm,cluster analysis is carried out on the CT enhanced image data of individual patients and the characteristic matrix of three basic statistics is extracted.Fourthly,logistic regression analysis algorithm and Adaboost algorithm were used to binary classify the eigenmatrices extracted by the above four clustering algorithms,and the accuracy,confusion matrix and ROC curve were compared.The results show that the median features obtained by the deep clustering algorithm based on variational autoencoder have the best predictive effect based on the enhanced CT image data of stage II colon cancer patients.In terms of processing the unbalanced CT value data of the colon cancer patients,the accuracy rate of Adaboost algorithm is up to 92.31%,which is obviously better than logistic regression analysis.In addition,the AUC value is 0.94.In conclusion,by predicting whether cancer metastasis occurs in colon cancer patients,this study found that the prediction accuracy of data in phase II patients was higher.This increases the chance of early detection of tumor metastasis and provides new ideas for the treatment and prognosis of colon cancer patients.