Radiomics To Predict Recurrence Outcomes And Prognosis Of Head And Neck Cancer

The primary sites and pathological types of head and neck cancer(HNC)rank first in all kinds of tumors of the whole body.While in the treatment of HNC,the recurrence of the primary tumor prediction is great significance for doctors to map out a therapy plan.Radiomics use radio methods to evaluate heterogeneity of tumor tissue level,and considering clinical information correlation,the study can promote the implementation of personalized precision treatment.This article retrospectively analyzed the HNC data published by the TCIA database with 298 HNC patients’ raw PET/CT image data included some clinical information.Medical fusion method based on wavelet transform used to fuse PET/CT images,and extracts 3,215 tumor features,including first-order gray information(10),3D shape features(5),and 3200 texture features.Firstly,logical regression model,Cox regression model,and K-M survival analysis used to evaluate whether there were difference in the predictive performance of the local recurrence and distant metastasis using fusion images.Secondly,the clinical information was added to the Radiomics model.Random forest prediction model was established including all feathers chosen.Finally,distant metastasis of HNC was predicted and classified using convolutional neural networks in deep learning.We found that in the prognosis of local recurrence,the fused PET/CT feature set do have better performance.Either consistency index CI 0.63(from 0.52 to 0.63)or the AUC0.60(from 0.57 to 0.60)got the best result in all four data sets when the clinic information were not included.After adding clinical information,compared with non-fusion,the consistency index increased from 0.67 to 0.70,and the AUC value increased from 0.69 to0.70.Using PETs,the deep learning-based Radiomics perform well in predicting distal metastasis which got sensitivity 0.67 and specificity 0.72.The results of this article show that the multi-modal fusion image preprocessing method can make the radiomics features extraction more complete.Radiomics can provide supplementary information for the design of clinical trials and ultimately help guide the precise treatment of head and neck cancer.