Research On Aided Diagnosis Of Dementia Based On Image Analysis

With the global trend of aging,dementia has affected millions of people in the world and the number of patients is still growing rapidly.Mild cognitive impairment(MCI)and Alzheimer’s disease(AD)are both prodromal symptoms leading to dementia.Therefore,it is necessary to distinguish MCI,AD from healthy middle-aged and elderly people.Studies have shown that people with dementia have drawing difficulties.Complicated drawing tasks(e.g.,the tree-drawing test)allow for better observation of drawing impairment than figure reproduction.Previous studies have used the tree drawing test to screen for cognitive impairment,but few people have focused and comprehensively paid attention to the details of tree drawing.In the research of this thesis,a total of 81 subjects were recruited and their tree drawing data were collected through digital boards.Starting from the two feature dimensions of quantitative and qualitative,51 features were extracted from the details and the whole of tree paintings,respectively.Then,5 sets of binary classification experiments were carried out using widely used machine learning models,and additional experiments were performed to eliminate the interference of gender,education,and age.Then this paper ranks the importance of the features selected by the model.It can be seen from the machine learning classification results that the binary classification of different cognitive situations has good performance,and the features extracted in this paper have played a great role.This preliminary study suggests that the treedrawing test has the potential to be used to develop machine learning tools to support the diagnosis of cognitive impairment diseases.The innovations and main work of the paper are as follows:(1)Focus on the characteristics of the painting tree itself and enrich the indicatorsSome literatures use drawing software and digital boards to collect test data of tree drawing,and focus on exploring the relationship between quantitative features such as line and color changes,gray-level cooccurrence matrix and cognitive impairment.Another part of the literature sees the importance of tree painting drawing conditions,extracting features from both quantitative and qualitative dimensions.This final project integrates previous research,further enriches the qualitative features,evaluates the details of the tree painting more comprehensively,and also includes the quantitative data of the digital board.While obtaining better classification results,the validity of the new features for classification results is verified.(2)Explore the feasibility of machine learning classification methodsPrevious studies all used statistical methods and did not conduct comparative experiments on different classification models.This final project selects the machine learning classification scheme widely used in the literature to explore the feasibility of the classification method and provide theoretical support for automatic scoring using machine learning.To evaluate the effectiveness of the above classifiers,different evaluation metrics are output.It can be seen from the classification results that each binary classification experiment has a good model performance,and even reached an accuracy rate of 90%in some experiments.(3)Provide feature importance rankings for different two categoriesPrevious studies only explored the predictors of categorical cognitive impairment,but did not discuss the importance of each feature for categorization.This final project outputs the importance of all selected features.Rank the importance of the outputs of three machine learning models under the same classification experiment.Finally,the top 10 features ranked by importance under each experiment are output.This thesis once again verifies the importance of some features in previous studies for distinguishing different levels of cognition,and at the same time proves that the new features included in this paper also play a greater role in the classification process.