Construction Of The Rectal Cancer Radiomic Signature And The Disease Free Survival Analysis

Currently,Neoadjuvant chemoradiotherapy(n CRT)followed by total mesorectal excision(TME)is currently considered the standard combined modality treatment for patients with locally advanced rectal cancer(LARC).This therapeutic strategy improves local control of the disease,but does not notably increase overall survival(OS)or diseasefree survival(DFS).The reported 3-and 5-year cumulative incidence rates of distant metastasis after n CRT are 22% and 25%,respectively.Meanwhile,distant metastasis is the main cause of treatment failure in patients with LARC who undergo n CRT.In individuals at high risk for an adverse outcome after n CRT,additional systemic therapy may reduce the risk of distant relapse,conferring a survival benefit.Therefore,identifying adverse prognostic features that affect survival and preoperative risk stratification can help select individualized treatment plan and improve the prognosis of patients with LARC.Radiomics is an emerging field that converts imaging data into a high dimensional mineable feature space using a large number of automatically extracted datacharacterization algorithms.The imaging features capture distinct phenotypic differences of tumours and may have prognostic power.Tumors are heterogeneous both in space and time,which is a clear barrier to the goal of personalized therapy based on molecular biopsy-based assays,and the biopsy has a very heavy burden on the patient.Based on this situation,there is an urgent need for effective excavation of image information.This poses a major opportunity and challenge for radiomics.The significant advantage of radiomics is that it can evaluate the internal heterogeneity of the tumor based on radiological examination before surgery to quantify the tumor microenvironment.In this thesis,we use radiomic method to construct a predefined radiomic quantified feature library to quantitatively describe tumor heterogeneity.Moreover,the traditional machine learning model was combined to construct radiomic signature and create a scientific,quantitative and simple prognostic analysis model for rectal cancer.The model can predict and analyze the prognosis of rectal cancer patients before surgery.Detailed studies and innovation works are as follows1.First of all,this thesis transforms the MR images to a similar gray range at preprocess step and then expands the traditional radiomic features.The 485 radiomic features are divided into 4 groups:(I)tumor intensity,(II)shape and size,(III)texture,and(IV)wavelet characteristics.The radiomic quantified feature library was constructed based on these features.2.The prognosis of two groups of patients was compared in the traditional prognostic studies based on texture analysis.The mining of image is very simple.In this thesis,we use the radiomic quantified feature library to build a stacked machine learning model with better predictive power than a single model.It can establish a mapping relationship between features and the survival outcomes of patients and then construct a radiomic signature.The radiomic signature combined with clinical risk factors can improve the predictive ability of the model.3.The model is difficult for clinicians to understand.This is the biggest problem faced by the model.This thesis simplifies the survival analysis model to directly obtain the probability of occurrence of the terminal event for each individual,generating a friendly graphical interface(nomogram)to assist the doctor in clinical diagnosis.