Radiomics-based Predictive Research Of Progression-free Survival In Tumors

The incidence and mortality of cancer have been high rate all over the world,seriously threatening the health of humans.Early prognosis predictions are critical for cancer patients,and the lack of reliable predictors of prognosis is one of the vital causes for high mortality rates in patients with tumors.Medical imaging,such as magnetic resonance imaging(MRI),computed tomography(CT),etc.,are routine tumor assessment tools,which rely on the manual reading of radiologists to give qualitative assessment of the tumor response.In recent years,with the development of artificial intelligence and radiomics,a new idea has been proposed by researchers to study tumors’ prognosis,making MRI,CT and other medical images as reliable,accurate and quantitative prognostic assessment tools possible.Based on artificial intelligence and medical image big data,radiomics has been successful applied in predicting the prognosis of some tumors.However,there is still a shortage of advanced high-grade serous ovarian cancer(HGSOC)and skull base chordoma(SBC)around the world.To address this problem,three aspects were conducted in this study:(1)A radiomics-based prediction method for progression-free survival(PFS)of tumors was developed;(2)The quantitative prediction method for PFS based on preoperative contrast enhanced CT images of advanced HGSOC patients was realized;(3)The quantitative prediction method for PFS based on preoperative routine MRI images of SBC patients Realized.The main innovations and contributions of this study are as follows:1.A radiomics-based prediction method for PFS of tumors was developed.The medical imaging data of the current study was obtained from the Picture Archiving and Communication System(PACS)of the collaborative hospital.The data format was Digital Imaging and Communications in Medicine(DICOM).The region of interest(ROI)was acquired from manual segmentation by radiologists.The ROI was manual segmented by three radiologists using ITK-SNAP(www.itksnap.org)software.Two of the radiologists drawn the tumor area on the patient’s medical image,and another senior radiologist determined the final ROI.Intra-class Correlation Coefficient(ICC)was applied to verify the stability of radiomic features extracted by the ROI of two radiologists.Stability of radiomic features was defined as ICC greater than 0.75.The segmentation files were stored in the mha format.The nearest neighbor interpolation algorithm was applied to normalize the original images and the ROI of the tumors to eliminate the impact of inconsistent scanner and parameters(Slice Thickness,Pixel Spacing,etc.)on radiomic features.Based on the studies from Hugo J.W.L.Aerts and previous studies,620 quantitative radiomic features were extracted in the present study,including 17 histogram features,8 shapes features,51 texture features,and 544 wavelet features.Univariate feature analysis and Elastic Net were used for dimensionality reduction,feature selection,and radiomic signature building.Kaplan-Meier survival analysis and Log-rank test were used for determining the relationship between prognostic factors and PFS.Univariate and multivariate Cox proportional hazards model were applied to calculate the Harrell’s concordance index(C-index)and predict the individual probability of PFS at critical time points.Time-dependent C-index was applied to evaluate the discriminant accuracies of the univariate Cox model.De Long’s test was applied to compare the predictive performance of each model.The 95% confidence interval(CI)for C-index was calculated by 1000 re-sampling.The calibration curve was applied to estimate the degree of variation predicted by the predictive model,and its prediction probability was compared with the true probability.The calibration curve was verified by using the HosmerLemeshow test.All statistical tests were two-sided and P < 0.05 was considered significant.2.A radiomics-based PFS prediction method for patients with advanced HGSOC was realized.In the retrospective study of PFS prediction in patients with advanced HGSOC,142 patients with advanced HGSOC were enrolled.All enrolled patients were followed-up for at least 18 months,with a follow-up time of 18.8-81.8 months and a median follow-up time of 38.8 months.Patients were divided into three groups based on the time of surgery and the location of the hospital: the training cohort,the internal validation cohort from West China Second University Hospital,Sichuan University,and the independent external validation cohort from Henan Provincial People’s Hospital.The patients’ preoperative abdominal pelvis contrast enhanced CT images data was used for radiomic analysis.The radiomic signature was built by four radiomic features selected by the LASSO(Least Absolute Shrinkage and Selection Operator)regression.Patients of three cohorts were successfully divided into two groups(high/low risk)with statistically significant differences respectively.Furthermore,the discriminative accuracies of radiomic signature and radiomic nomogram for predicting disease progression risk within 18 months and 3 years were higher and significantly better than the clinical model.3.A radiomics-based PFS prediction method for patients with SBC was realized.In the retrospective study of PFS prediction for patients with SBC,148 SBC patients were enrolled from Beijing Tiantan Hospital,Capital Medical University,of whom 64 had disease progression.The follow-up time was 4-122 months and the median follow-up time was 52 months.Patients were divided into training and validation cohorts based on the time of surgery.The preoperative axis T1 FLAIR(Fluid Attenuated Inversion Recovery),T2 weighted imaging,and enhanced T1 FLAIR MRI sequences were used for radiomic analysis.A total of 1,860 3-D radiomic features were extracted from the three sequences(620 of each sequence,respectively).The radiomic signature was built by 18 radiomic features and normalized to categorical variable(high risk,medium risk,and low risk)and continuous variable(between 0-1).The categorical radiomic signature successfully divided the patients of validation cohort into high-risk,medium-risk,and low-risk groups.And the PFS in different groups had statistically significant difference.The C-index and discriminative accuracy of predicting 5-year progression risk of the continuous radiomic signature were higher and better than nine other potential clinical prognostic factors.In all,radiomics can be used for prognostic evaluation of tumor patients,and radiomic signature is a quantitative and reliable prognostic marker for tumors.Radiomics is a lowcost,non-invasive method to evaluate the prognosis during perioperative period,which can not only directly affect the formulation of clinical treatment and follow-up plan,but also has positive significance to realize individualized treatment and prolong the survival of patients,and is an important supplement to precision medicine.