MRI data for early detection of Alzheimer diseases

Alzheimer’s disease is a neurologic disorder caused by abnormal buildup of proteins in brain cells, which eventually lead to the brain to shrink and cell death. This is a progressive disease which continually declines thinking capacity, behavioral and social skills. According to World Alzheimer Report 2018, there is currently 50 million people worldwide living with Alzheimer till the date of the study. The number has been surely increased afterwards [1]. Currently there is no cure for Alzheimer’s diseases. However, the available medicines can help to reduce the symptoms in temporary manner [2]. 

Now the researchers are trying to focus on early detection of Alzheimer, which would help to fight back with symptoms in earlier stages and protect the cognitive skills from getting lost. 

Alzheimer’s Diseases Neuroimaging Initiative (ADNI) published some data set for public access related to Alzheimer’s diseases. There is a Magnetic Resonance Imaging (MRI) dataset available of 150 subjects aged from 60 to 96. Among them, 64 subjects were detected with dementia and 72 were healthy. Moreover 14 subjects were non-demented in first visit, while they were detected with dementia in a later visit. Some standard tests have been performed for detection while MRI was being acquired. 

Mini Mental State Examination (MMSE):

It is a series of questions and tests having points for each one which is used to determine individual’s mental ability, memory, attention and language. It is a very important assessment technique of dementia. 

Clinical Dementia Rating (CDR):

This is rating based test performed to detect the severity of dementia. The parameters for testing are Memory, Orientation, Judgment and Problem Solving, Community Affairs, Home and hobbies and Personal Care. The scale is from 0 to 3, where none = 0, questionable = 0.5, mild = 1, moderate=2, severe = 3.

Estimated Total Intracranial Volume (eTIV):

Intracranial volume (ICV) is an important normalization measure used in morphometric analyses to correct for head size in studies of Alzheimer Disease (AD). Inaccurate ICV estimation could introduce bias in the outcome.

Normalized Whole Brain Volume (nWBV):

This is MRI based volumetric measurement used to detect the progression of atrophy. T1- weighted imaging is used for the study. This is a simple protocol for measuring TIV from T1-weighted MR images, and to apply TIV normalization to serial brain measures in controls and subjects with Alzheimer disease (AD).

Atlas Scaling Factor (ASF):

This is also a volume-scaling factor used for detection of Alzheimer in spatial way. Automated atlas transformation generated the Atlas Scaling Factor (ASF) defined as the volume-scaling factor required matching each individual to the atlas target. Because atlas normalization equates to head size, the ASF should be proportional to TIV.

Machine learning models are used to train the dataset with the specified features. Other important features considered, are Years of Education (EDC), Socioeconomic status (SES), number of visit and MR delay. In I3TK, two models have been designed, for both of them; received accuracy and AUC are quite significant, which is better than most of the previous result statistically.  Multi Linear Perceptron (MLP) and Random Forest (RF) algorithms have been used, and both provided almost same result. MLP model shows accuracy of 91.96% and 5 fold validation scores are 92.45%, 98.07%, 96.15%, 96.15%, and 92.30%. For RF, these values are 92.45%, 98.07%, 96.15%, 96.15%, and 92.30%.

In the table below (Table 1), all the previous works along with the current work have been shown, 

Table 1

Sr.No.PaperDataModelResults
1.E. Moradi et al. [3]Ye et al. [7]Random Forrest ClassifierAUC = 71.0%ACC = 55.3%
Filipovych et al. [8]Random Forrest ClassifierAUC = 61.0%ACC = N/A
Zhang et al. [9]Random Forrest ClassifierAUC = 94.6%ACC = N/A
Batmanghelich et al. [10]Random Forrest ClassifierAUC = 61.5%ACC = N/A
2.Zhang et al. [4]Ardekani et al. [11]Support Vector Machine
polynomial kernelAUC = N/AACC = 92.4%
linear kernelAUC = N/AACC = 91.5%
radial basis functionAUC = N/AACC = 86.7%
3.Hyun, Kyuri, SaurinMarcus et al. [1]Logistic Regression (w/ imputation)AUC = 79.2%ACC = 78.9%
Logistic Regression (w/ dropna)AUC = 70.0%ACC = 75.0%
Support Vector MachineAUC = 82.2%ACC = 81.6%
Decision Tree ClassifierAUC = 82.5%ACC = 81.6%
Random Forest ClassifierAUC = 84.4%ACC = 84.2%
AdaBoostAUC = 82.5%ACC = 84.2%
4.Hyunseok Choi,MS, University of PittsburghLogistic Regression(dropna)AUC=79.1%ACC = 75%
SVMAUC = 82.22%ACC = 81.5%
Decision TreeAUC = 82.5%ACC = 81.57%
Random ForestAUC = 84.44 %ACC = 84.2%
AdaBoostAUC = 82.5%ACC = 84.2%
5.ShuvodeepSaha,
Researcher, International Institute of Innovation and Technology, Kolkata
Multi-Layer PerceptronAUC = 92.5%ACC = 91.96%
Random ForestAUC = 92.5%ACC = 91.07%

The importance of feature is also observed to identify which is most effective among all to detect the possibility of dementia in individuals. Random Forest algorithm is used for understanding the weightage distribution among the features. CDR and MMSE show the maximum importance and thus, these two features can be considered as most significant (see Figure 1). 

Figure 1: Feature wise importance

Therefore, the results indicates that imaging technique can be used in the future for the early detection of dementia, with larger sample size, that will help to detect the Alzheimer’s symptoms in the stage where the degradation of the cognitive function is not that significant and reversible. Moreover, the study should be done with a more versatile and wide range of dataset and populations all over the world.

Project Link: https://www.kaggle.com/deep07/mri-data-for-alzheimers-prediction

References:

[1] https://www.alzint.org/u/WorldAlzheimerReport2018.pdf.

[2] https://www.nhs.uk/conditions/alzheimers-disease/treatment/

[3] https://www.alzheimers.org.uk/about-dementia/symptoms-and-diagnosis/diagnosis/mmse-test

[4] https://pubmed.ncbi.nlm.nih.gov/8232972/

[5] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4423585/

[6] https://pubmed.ncbi.nlm.nih.gov/11559495/

[7] https://pubmed.ncbi.nlm.nih.gov/15488422/

Leave a Comment