| The dementia disease affects a person every few seconds in the world.The total number will increase every year.It indicates a decreased brain function.Many studies help dementia patients to use image research and non-image research.We have done two things: dementia v.s.non-dementia classification and dementia-level classification.There are two outputs in the dementia v.s.non-dementia classification: dementia and non-dementia.After we finish classify dementia v.s.non-dementia,we continue to classify the dementia-level to know the level of someone who has dementia.There is two outputs in dementia-level classifications: low-level and high level.We use a perceptron neural network and an OASIS data-set for our research.We develop the preprocessing technique to minimize the high dimensional data and reduce the noise of data.We have 10 features: age,sex,hand,education,SES,MMSE,ASF,e TIV,n WBV and target feature used CDR.We use the technique of 10-fold Cross-Validation and the confusion matrix to illustrate the accuracy.From these results,we compare a perceptron with SVM algorithm and compare it with another study.For the dementia v.s.non-dementia classification,the mean accuracy of each process is 96.56%,the mean of sensitivity is 98.7% and the mean of specificity is 94.5% in a perceptron process.In the SVM process,the mean accuracy of each process is 90.28%,the mean of sensitivity is 90.66% and the mean of specificity is 90%.For the dementia-level classification,the mean accuracy of each process is 94.67%,the mean of sensitivity is 96.2% and the mean of specificity is 93.4%.The results from SVM,the mean accuracy of each process is 91.00%,the mean of sensitivity is 91.5% and the mean of specificity is 90.5%.From the results,it can be seen that the perceptron is good to do binary classification such as this study: dementia v.s.non-dementia and dementia-levels.However,the perceptron is more precise for its accuracy and it is very efficient for binary classification. |
PDF Full Text Request