A Radiomics-Based Study For Preoperative Assessment Of The Survival Risk In Astrocytoma And Intrahepatic Cholangiocarcinoma

To discover the cure for cancer,many novel treatments have been developed to reduce the mortality rate of cancer year by year.However,due to the individual inherent differences of patients,the prognosis of patients receiving the same treatment was found to vary widely.Therefore,it is essential to non-invasively assess the cancer survival risk before treatment,which could assist clinicians to make personalized decisions and surveillance.In clinical practice,clinicians assess the cancer survival risk based on the clinical manifestations of patients.However,this method is dependent on the experience of the clinician and not be able to quantify.A newly emerging techniqueradiomics provides a potential approach to solve this targeted clinical issue.Based on machine learning methods,radiomics harnesses mineable quantitative features extracted from encrypted medical images along with clinical or genetic data to produce an evidence-based clinical decision support system.By comprehensively characterizing tumor heterogeneity and invasiveness,radiomics is expected to achieve personalized quantitative preoperative survival risk assessment.In this study,we explored the application of radiomics analysis in cancer survival risk assessment about astrocytoma and intrahepatic cholangiocarcinoma.For astrocytoma patients,temozolomide(TMZ)chemotherapy is a noteworthy adjuvant therapy,and the prognosis after chemotherapy is highly related to Oxygen 6methylguanine-DNA Methyltransferase(MGMT)status.Since the follow-up data for the patients receiving TMZ chemotherapy are hard to collect,machine learning algorithms cannot be directly applied to make the prognosis prediction.Therefore,this study aimed to indirectly achieve survival risk assessment of patients receiving TMZ chemotherapy via predicting MGMT status.First,edema and tumor regions were manually segmented from multi-sequence magnetic resonance(MR)images of astrocytoma.Second,handcrafted radiomic features were extracted from the region of interest.Third,six single-sequence radiomics signatures were constructed by logistic regression after feature selection.Last,the signatures were combined into a fusion radiomics signature.The results showed that multi-sequence and multi-region fusion radiomics signature achieved the best performance,and successfully stratified progression risk in the low-risk and high-risk groups of patients receiving TMZ chemotherapy.For patients with intrahepatic cholangiocarcinoma,the high postoperative recurrence rate leads to a poor prognosis.Therefore,quantitative assessment of recurrence risk before surgery is essential for formulating more effective treatment and follow-up strategies.Due to the low incidence,intrahepatic cholangiocarcinoma is difficult to collect a lot of follow-up data.Only a few radiomics studies with a single center and small samples made preliminary explorations about the early recurrence prediction of intrahepatic cholangiocarcinoma.This study is the first multi-center radiomics study on this clinical issue.By extracting radiomic features from the tumor region of computed tomography(CT)images,a variety of feature selection algorithms and classifiers are compared to identify the optimal radiomics signature.The results suggested that the radiomics signature quantitatively predicts the early recurrence risk of patients with intrahepatic cholangiocarcinoma,and significantly outperforms the clinical model.The two abovementioned studies confirm that radiomics could quantitatively assess the survival risk in astrocytoma and intrahepatic cholangiocarcinoma,which may provide a quantitative basis and clinical assistance for clinical decision-making.