This project focuses on the early prediction and diagnosis of radiation-induced brain injury in nasopharyngeal carcinoma patients. Based on the big data of imaging and serum metabonomics samples, combined with the machine learning analysis method, dynamic evolution mode of radio-metabolomics characteristics was analyzed . The potential internal relationship between brain structure and serum metabolic changes was explored, and the individualized prediction model was constructed to screen out the high-risk patients with brain injury after tumor radiotherapy, so as to provide reference for the diagnosis of radiation-induced brain injury caused by tumor. radiotherapy Intelligent diagnosis provides a new theoretical and practical basis.
Research Process 1. The MRI based cohort data set of nasopharyngeal carcinoma was established, and the data of multiple follow-up time points before and after radiotherapy (including initial diagnosis, 6 months, 12 months and 24 months after radiotherapy) were standardized to obtain the longitudinal data set; 2. Region of interest (ROI): it mainly delineates the bilateral temporal lobe, brain stem and other brain regions, and extracts the corresponding image features in ROI; 3. Feature selection: using the strategy of radiomics combined with Artificial Neural Network to reduce the dimension of high-dimensional image features, the key features are selected and used for the subsequent construction of classification and prediction model; 4. Extracting key features: using vertical axis data analysis method and logistic regression to establish dynamic prediction model.
Study Type
OBSERVATIONAL
Enrollment
800
The patients got intensity-modulated radiation therapy during observation
Xiangya Hospital of Central South University
Changsha, Hunan, China
RECRUITINGThe variations in imaging features from initial diagnosis to 24 months after radiotherapy
Brain MRI image data of included patients with MRI sequence (T1 Wi, T2 Wi, T1 + C etc.) before and after radiotherapy (including initial diagnosis, 6 months, 12 months and 24 months after radiotherapy) were obtained for Artificial Neural Network analysis. The key features were found out by machine learning.The variations in imaging features from initial diagnosis to 24 months after radiotherapy were abtained to conduct an efficient prediction model for the probability of radiation encephalopathy.
Time frame: Before and after radiotherapy (including initial diagnosis, 6 months, 12 months and 24 months after radiotherapy). All the data got from each time point were used to conduct an efficient prediction model.
Changes in metabolic feature from initial diagnosis to 24 months after radiotherapy
Since the changes in serum nucleotide metabolism, amino acid metabolism, fat metabolism were observed in the radiation encephalopathy patients. All Gas chromatography-mass spectrometer(GC-MS) data including retention features, peak intensity and integral mass spectrometry for each serum sample are used for analysis, to predict whether the separation between the radiation encephalopathy patients group and the control group is significant. The serum metabolism changes of patients during two years after radiotherapy are followed to obtain metabolic footprint. The serum sample got from different time points were applied in agglomerate hierarchical clustering for the screening and identification of various metabolites in the serum samples to get biomarkers, which can evaluate the changes of the metabolites in radiation encephalopathy.The PLS-DA model is used to represent changes in metabolic feature during metabolism.
Time frame: Before and after radiotherapy (including initial diagnosis, 6 months, 12 months and 24 months after radiotherapy). The PLS-DA model is used to represent changes in metabolic feature during metabolism.
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.