Application Research Of Radiomics In Radiotherapy Of Esophageal Cancer And Head And Neck Cancer

Malignant tumors are an important factor threatening human health due to their high morbidity and high mortality.As a local treatment method,radiotherapy plays an increasingly important role in the clinical treatment of tumors.It aims to raise the local contral probability of the lesion by increasing the dose on tumor target and reducing radiation induced injury of surrounding normal tissue,in order to improve the survival rate and overall quality of life for the patients.However,although the latest Intensity Modulated Radiation Therapy(IMRT)effectively improves the irradiation accuracy and uniformity of the tumor target,the treatment is very complicated and the curative effect is difficult to predict,due to the heterogeneity and the complex microenvironment of the malignant tumor.Moreover,the target lesion is normally surrounded by normal tissues,which are inevitable to be irradiated by X-ray and presented of irradiation toxicity.Individualized treatment has been gradually recognized for different pathological and structural characteristics of tumors.A large number of literatures indicate that early prediction of tumor treatment response and side effects is conducive to the treatment and recovery of patients,further guiding tumor treatment and improving the survival rate of patients.Therefore,clinical practice requires a predictive technique for the efficacy and complications of radiotherapy,which can predict the efficacy and the risk of complication before treatment,thus optimizing the patients’ treatment plan.Radiomics is an emerging field to decode tumor phenotype by quantitative analysis of image features extracted from tumor volume.Before radiation treatment,a large number of images are needed for target delineation and planning design,which provides a good opportunity and theoretical basis for the combination of radiomics and radiotherapy.The thesis starts with the pre-treatment medical imaging of esophageal cancer(EC)and head and neck cancer(HNC)patients and is devoted to studying the radiomics predictive model,using statistical analysis and machine learning method,for early prediction of response for EC and radioresistance for HNC recurrence.In addition,we explore the value of deep learning-based radiomics method in the prediction of radiation-induced lung injury.Several research achievements and contributions are summarized as follows:1.Radiomics in prediction of treatment response to chemoradiotherapy in esophageal carcinomaa)To summarize the existing four radiomics feature extraction algorithm from literatures and analyze their mathematical and radiological significance;b)To explore the value of pre-treatment contrast enhanced-CT based radiomic features in prediction of treatment response for EC and to build predictive models using machine learning method;c)To further explore the value of T2 W and SPAIR T2 W MR based radiomic models in prediction of treatment response for EC and to compare the predictive performance between the two sequences.Then analyze its principle and clinical significance.2.Radiomics in the recurrence patterns of head and neck cancera)To analyze the recurrence patterns and reasons in patients with head and neck cancer(HNC)and to investigate the relationship between in-field recurrence and radioresistance;b)To analyze the recurrence patterns of nasopharyngeal carcinoma(NPC)patients in our centre by using dose-accumulation tools and screen out the patients who have in-field relapsed due to radioresistance;c)To propose a strategy originally: to extract quantitative features from pre-treatment SPAIR T2 W MR image of the patients,and then combined with machine learning method to establish predictive models for radioresistance.3.The application of deep learning-based radiomic in prediction of radiation-induced lung injury(RILI).a)To summarize the progress of CT-based radiomics in prediction of RILI,and further focus on the prediction of local RILI by using pre-theraputic CT of high-dose area;b)To propose transfer-based deep convolutional neural network(CNN)model,named deep learning-based radiomics,which enables to autonomously extract features related to RILI,and use the fine-tuned model(AlexNet and GoogleNet)to predict RILI;c)To test the prediction ability of the pre-trained AlexNet and GoogleNet on local RILI after limited medical instances fine-tuning,and analyze its principle and clinical significance.