| Head and neck cancer refers to tumours that develop in the head and neck,including laryngeal,oropharyngeal,nasopharyngeal and hypopharyngeal cancers.Over 90% of patients have squamous cell carcinoma,which has the highest number of primary sites and pathological types of tumours in the whole body.Accurate prognostic prediction is important for head and neck tumours.However,due to tumour heterogeneity,a single diagnostic pathological biopsy cannot accurately and comprehensively characterise the dynamic changes of tumours;although multi-regional biopsy can improve the ability to determine the degree of spatial heterogeneity of tumours,it is risky and has not been widely used.Therefore,there is an urgent need to develop a new prognostic tool to assist clinical practitioners in diagnosis and treatment.Currently,imaging histology has demonstrated its potential value in the clinical setting with its non-invasive detection of medical images and comprehensive portrayal of tumour information.Also,as one of the main treatment modalities for tumours,the dose distribution in radiotherapy provides a wealth of prognostic information.In view of this,in this study,a multimodal,multiparametric imaging histology prognostic model was constructed by combining conventional imaging histology as well as dose distributions during radiotherapy,and focusing on the role of the model in predicting overall survival of head and neck cancer patients.This study used 220 patients with pathologically confirmed head and neck cancer from four centres from the TCIA public database as the study,and PET,CT and 3D dose distribution(RTdose)images were acquired for each case.The cases were divided into three different training and test sets(CEN12 vs CEN34,CEN23 vs CEN14,CEN13 vs CEN24)based on centre two-by-two to validate the predictive performance of the same model under different dataset divisions.For each case,2260 PET/CT imaging histology features,1116 dosimetric features and 8 DVH parameters were extracted and all features were normalised.The features of each modality were filtered by univariate analysis and correlation coefficient statistics,and combined into seven different groupings of PET,CT,Dose,PET+CT,PET+Dose,CT+Dose and PET+CT+Dose in turn.The features of the different groupings were finally brought into the LASSO-Cox model,and the predictive performance of the model was assessed using the C-index as well as the Kaplan-Merier curve.A Nomogram was also used to visualise the relationship between dose histological features,imaging histological features and overall patient survival in the model.The experimental results show that the predictive performance of the model for PET+CT is significantly better than that of the single PET or CT imaging histology model in the conventional imaging histology model,even under different training set and test set divisions.The C-index of the test set for the PET+CT model under different data set divisions was0.846,0.762 and 0.815,the C-index of the test set under different divisions for PET was0.781,0.761 and 0.79,and the C-index of the test set under different divisions for CT was0.787,0.668 and 0.71.In incorporation of dose features(including dose histology features and DVH parameters),the predictive performance of each model was improved to some extent.The C-index of the test set under different divisions for the PET + CT + Dose model was 0.873,0.759 and 0.835,the C-index of the test set under different divisions for the PET+ Dose model was 0.824,0.756 and 0.777,and the C-index of the test set under different divisions for the CT + Dose model was In the Kaplan-Merier curves for the different groupings,the CT-only model failed to classify patients into high and low risk groups under CEN12 vs CEN34 and CEN13 vs CEN24(Log-rank test,p>0.05),and the PET+CT model failed to classify patients into high and low risk groups under CEN23 vs CEN14(Log-rank test,p>0.05).vs CEN14 did not classify patients into high and low risk groups(Log-rank test,p>0.05).The results of this study demonstrate that traditional radiomics,combined with dosiomics features and DVH parameters,can have a positive effect on the prediction of overall survival in patients with head and neck cancer and can well classify patients into highand low-risk groups(Log-rank test,p < 0.05). |
