Progression-Free Survival Of Nasopharyngeal Carcinoma Based On Dosiomics And Radiomics

Nasopharyngeal carcinoma(NPC)is a common malignant tumor originating from nasopharyngeal mucosa,with a high incidence in southern China.Intensity modulated radiation therapy(IMRT)has been widely used in the treatment of NPC and shown promising effects.The dose distribution in the radiotherapy target area of NPC is closely related to tumor progression.It has the potential to be used as a marker for diagnosis and prognosis.Dose volume histogram(DVH)is mainly used in clinical practice to evaluate the target exposure dose but it can only provide two-dimensional dose-volume statistical information.Inspired by radiomics,Dosiomics came into being.Dosiomics describes the dose distribution in three-dimensional space.Aims to obtain additional information of diagnostic value,dosiomics quantifies subtle variations in spatial dose distributions by calculating quantitative features.Dositomics has been preliminarily studied in the diagnosis and prognosis of many types of cancer,but there is little research of the NPC prognosis.Thus,this study aims to investigate the prognostic value of dosiomics combined with radiomics in NPC.Two outcomes,recurrence-free survival(RFS)and metastasis-free survival(MFS)were considered in this study.The main contents are as follows:1)CT,MR and dose images of 172 NPC patients were collected,and machine learning and integrated learning algorithm were used to construct the prognostic models.The results show that the performance of multimodal models is significantly higher than that of single modality models(p<0.01).The XGBoost model with the combination of CT,MR and dose features reached the best prognostic performance(RFS:AUC＝0.80±0.06,PR-AUC=0.69±0.12;MFS:AUC=0.85±0.07,PR-AUC=0.73 ± 0.12).2)Investigated the prognostic performance of multi-level CT-dose fusion dosiomics at the image-,matrix-,and feature-levels from the gross tumor volume at nasopharynx(GTVnx),the gross tumor volume at the involved lymph node(GTVnd)and the peritumoral region(3 mm and 5mm)for NPC patients.The image level fusion models’ performance was better than the matrix/feature level models and single CT/dose models.The features from GTV were associated with tumor recurrence,while the features from GTVnd were more associated with distant metastasis.Image level fusion model GTV+RING 3mm nxnd GFF that combined the information from GTVnx,GTVnd and 3mm peritumoral regions GTV+RING＿3 mm＿nxnd achieved the highest C-index both in RFS and MFS predictions(RFS:0.822;MFS:0.786).