Brain Health Assessment Based On Deep Learning

The structure of the human brain changes with age.In addition to the structural changes caused by the normal aging process,excessive functional decline and neurodegenerative diseases may also lead to abnormal changes in the brain's morphological structure.As a biomarker,brain age can measure how old the brain is.Abnormal changes in the brain make the aging of the brain deviate from the normal track,resulting in the deviation of brain age from the biological age of the individual.This bias reflects the extent of abnormal changes in the brain.Magnetic resonance imaging of the brain has a very high spatial resolution,which can clearly show the pathological tissue and normal tissue,providing a possibility for the prediction of brain age.The methods of brain age prediction based on magnetic resonance imaging are mainly machine learning and deep learning.Traditional machine learning methods require manual feature extraction from magnetic resonance images,and then implement regression and prediction.The process of manual feature extraction is tedious and depends on feature extraction method.The prediction accuracy will be affected by the extracted features.With regard to deep learning,most of the existing brain age prediction models use three-dimensional convolution,which has a large number of parameters,high requirements on computer hardware,and a long training cycle.The specific work for solving above problems is as follows.On the one hand,aiming at the problems of large number of three-dimensional convolution parameters and long training period,we proposed a two-dimensional convolution neural network model based on the combination of attention mechanism and residual module.The model was set up for experiments on public data sets,and the optimal data slice range and residual weight of the model were obtained through specific experiments.Then the model was trained and the parameters were fine-tuned to make the performance of the model reach a better level in the validation set.Finally,the model achieved a good effect on the test set,which verified the effectiveness of combining the model's attention mechanism with the residual module.On the other hand,machine learning was used.The grey matter data was obtained by a more rigorous data preprocessing process through a series of operation like segmentation,registration and resampling processes.Two-dimensional gray matter image slices covering aging sensitive regions were directly selected as input features,and the performance of various machine learning algorithms was compared to get the optimal algorithm for brain age prediction.Then the model was used to predict the brain age of the patients and the difference between brain age and chronological age was obtained by subtraction.Then,the clinical scale and the difference were analyzed to explore the potential relationship between the difference and disease.The results showed that there was significant correlations between the difference and clinical scale scores.