In-depth learning questions about glioma imaging

MRI imaging is widely used in preoperative diagnosis of glioma stoma and postoperative monitoring of tumor progressionFrom a research perspective, MRI provides a standardized approach to establishing patient baselines, monitoring clinical progress, and designing treatment optionsBut the biological diversity of gliomas often complicates MRI imagingEven if there are several methods of colloma histological classification, it is not enough to solve complex MRI performance, and thus constitutes a "big data" problemMadeleine MShaver, of the Department of Radiology at the University of California, Irvine,and others have studied the challenge of solving big data through deep learning from artificial intelligence, and the results are published in the June 2019 issue of The Cancersthe resultsresearchers point out that machine deep learning methods are suitable for solving big data problemsThe data obtained by computer algorithm is analyzed and studied by computer algorithm, and the conclusion is finally drawn from the input variablesMachine learning has been used to train computers to use artificial intelligence for pattern recognition (Figure 1)The classical machine learning method has certain limitations, in contrast, the deep learning method does not have to pre-select the characteristics of the data, but through the high and low-level multi-level mode of learning to extract its characteristics, and then determine the appropriate patternand and information classificationIn addition, deep learning neural networks can clarify the relationship between data characteristics and abstract and advanced levelsCurrently, the most advanced method of image classification is to use convolutional neural networks (CNN) for in-depth learningCNN can simulate the animal visual cortex by applying artificial neural networks, setting multiple input levels to form areas of field of view that overlap with each neuronThe authors then used the results of deep learning to illustrate the detection indicators, performance characteristics and consequence prediction of gliomas, focusing on quantifying the burden of disease, determining the structure and genetic characteristics of tumors and surrounding tissues, and predicting prognosis based on imaging informationDiagram 1 Flowcharts of different machine learning components in different disciplines are increasingly complex from left to right The orange box indicates the trainable component deep learning methods require quantitative indicators of treatment measures, risk stratification and outcome prediction before and after surgery, and the structure and methods of the latest machine learning systems before and after GBM surgery and surgery postoperative monitoring is usually a two-dimensional measurement of MRI-enhanced imaging, and the volume analysis of tumor size can be used as a strong survival predictor compared to linear computational techniques A semi-automatic assessment of the volume of a brain tumor reduces variability between observers and is highly repeatable For a more accurate tumor size approximation, quantifying the entire tumor in 3D provides more accurate measurements (
Figure 2 ) a comparison of 11-week linear 1D measurements (A) and machine learning volume analysis (B) in a 64-year-old male,2 Figure A indicates that the maximum diameter of a recurrent tumor must be selectively determined in 2D images, and Figure B shows that the semi-automatic volume analysis method can accurately obtain the maximum diameter of a recurrent tumor the important problem facing deep learning is the lack of annotation data and the limited popularization of calculation methods, which hinder sedit-in stogit with clinical workflows Deep learning algorithms need proper training and require a large amount of high-quality, well-adated data However, data aggregation between medical institutions is difficult, and even if large data from multiple hospitals can be consolidated, the annotation process is time-consuming and requires a high level of expertise Deep learning can only be effectively trained by incorporating faster, higher-quality annotation data conclusions
authors believe that deep learning applications can provide an effective solution for radiology image analysis The continuous development of radiography can promote the development of deep learning technology for rapid image analysis The deep learning analysis of radiological imaging for gliomas, including many explorations, such as tumor heterogeneity, large number of tumor genotype identification, identification of progress or false progression, prediction of tumor classification and survival, etc., contributes to the treatment of glioma and the realization of precision medical management strategies.