Selected Compilation: Artificial Intelligence Deep Learning Analysis Of Multimodal Imaging Features To Predict Brain Tumors...

Glioma is a common primary malignant brain tumor, precise molecular type is very important for the diagnosis, treatment and identification of risk factors and prognosisLi Sun of the School of Innovation and Entrepreneurship at Southern University of Science and Technology in Shenzhen, China, and others proposed a deep learning method based on artificial intelligence imaging to automatically divide and predict the prognosis of glioma seisking tumorsFor molecular division, three different 3D convolutional neural networks (CNN) architecture are integrated to achieve robust predictive performance through most rules, effectively reducing deviations and improving accuracyIn order to accurately predict the survival, the authors extracted 4524 radiological features from tumor multimodal images, used decision trees and cross-validation to select effective features, and finally used random forest tree models to predict the overall survival rate of patientsThe results were published online in August 2019 in Frontiers in NeuroscienceResearch Methodology
Using the 2018 BraTS database, the study collected preoperative MRI imaging data from 285 glioma patients and asked a qualified radiologist to do a quantitative analysis of multimodal imaging characteristics in tumor areasHowever, due to the high heterogeneity of tumor appearance and shape, boundary blur and imaging artifacts, quantitative analysis is difficult, while the automatic segmentation of ai-intelligence deep learning has the advantages of speed, accuracy consistency and anti-fatiguethe study to design a new in-depth learning framework based on artificial intelligence, to analyze tumors and their sub-regions from multimodal MRI imaging, as well as to extract radiological features and clinical characteristics from tumor sub-regions with clear molecular subtypes for survival predictionThe following steps are included: first, using an integrated model including three different 3D convolutional neural network architectures to divide tumor sub-regions, obtaining a robust majority by voting, then using gradient-enhanced regression modeltraining data and ranking and cross-validation of radiological characteristics based on the importance of variance, and finally, using a random forest regression model to fit training data and predict the overall survival rate of patientsAfter perfecting the training, it was verified by 66 patients with brain tumorsThe prediction results of this integrated model rankfifth fifth in more than 60 models, but the prediction accuracy of the parting is 61%Conclusionfinally, the results show that the performance of the integrated model is better than that of a single model, which shows the effectiveness of the integration modelThis method reduces model deviation and improves performanceHowever, the specificity of the model is much higher than the sensitivity, and needs to be further improvedCopyright Notice the copyright of works published byOutside InformationAPP, including but not limited to text, pictures, videos, are owned by the sponsor/original author andof the God's Information, and no one may steal any content directly or indirectly by means of adaptation, cutting, reproduction, reproduction, recording, etcwithout the express authorization of theoutside informationWorks authorized by theOutside Informationshall be used within the scope of authorization, please indicate the source:theof the Outside InformationIf there is a violation,outside the informationwill reserve the right to further pursue the legal liability of the infringeroutside the informationwelcome individuals to forward and share the works published by this number.