Alzheimer’s disease is a non-reversible, non-curable, and progressive neurological disorder that induces the shrinkage and death of a specific neuronal population associated with memory formation and retention. It is a frequently occurring mental illness that occurs in about 60%–80% of cases of dementia. It is usually observed between people in the age group of 60 years and above. Depending upon the severity of symptoms the patients can be categorized in Cognitive Normal (CN), Mild Cognitive Impairment (MCI) and Alzheimer’s Disease (AD). Alzheimer’s disease is the last phase of the disease where the brain is severely damaged, and the patients are not able to live on their own. Radiomics is an approach to extracting a huge number of features from medical images with the help of data characterization algorithms. Here, 105 number of radiomic features are extracted and used to predict the alzhimer’s. This paper uses Support Vector Machine, K-Nearest Neighbour, Gaussian Naïve Bayes, eXtreme Gradient Boosting (XGBoost) and Random Forest to predict Alzheimer’s disease. The proposed random forest-based approach with the Radiomic features achieved an accuracy of 85%. This proposed approach also achieved 88% accuracy, 88% recall, 88% precision and 87% F1-score for AD
Currently, there are around 55 million people in the globe suffering from dementia, and annually about 10 million new dementia cases are recorded [
Magnetic Resonance Imaging does not introduce any instrument inside the body. It produces three-dimensional structural images. When frequent scanning is vital magnetic resonance imaging becomes ideal for the brain. MRI uses strong magnets that generate a heavy magnetic field. When the radiofrequency pulse is applied, protons get excited and start spinning and breaking the equilibrium, moving against the magnetic field [
In paper [
Approach/model | Dataset | Accuracy | Reference |
---|---|---|---|
SVM | Taken from Smt. Kashi Bai Navale Medical Hospital Pune | 95% | [ |
KNN | OASIS | 74.73 | [ |
SVM with PCA | T1 weighted ICBM template | 64% for 3D feature and 72% for 2D features | [ |
SVM | OASIS | 84.62% | [ |
SVM with RBF kernel | ADNI | 78% | [ |
Random forest | OASIS | 81.19% | [ |
SVM | ADNI | 80% | [ |
Random forest | OASIS | AUC values varied from 56.63% to 84.09% based on the subcortical brain region | [ |
SVM | OASIS | 75.51 | [ |
SVM with polynomial kernel and KNN | OASIS | GLRLM features with the KNN classifier gave an accuracy of 65.15%, GLCM features with KNN gave an accuracy of 74.79%. GLRLM features with SVM polynomial kernel gave an accuracy of 87.4% while GLCM with SVM gave 87.55% accuracy. | [ |
SVM with RBF kernel | Taken from Chinese PLA General Hospital | 86.75% for classifying AD |
[ |
logistic regression | T1-weighted MPRAGE mages from the Zhejiang Provincial People Hospital | 68.4% | [ |
k-nearest neighbors | OASIS | 86.6% | [ |
1-NN with RDA, PCA, and LDA | T1-weighted MRI scans from Xuanwu Hospital, Beijing | Varied 63.2% to 89.7%, depending on the region of interest | [ |
SVM with Gaussian kernel | ADNI, and AIBL | AUC value of 74% for ADNI and 83% for AIBL | [ |
SVM | ADNI | 73.95% for T4 and 86.56% for T3 | [ |
SVM with linear data analysis | ADNI, AIBL, and CADD | 63% | [ |
SVM with PCA | ADNI | 89% | [ |
diagonal quadratic discriminant with PCA | ADNI | AUC for CN |
[ |
Logistic regression | ADNI | 79% | [ |
Support Vector Machine-based method with T1weighted MRI images | OASIS | 80.76% | [ |
Backpropagation network | OASIS | 78% | [ |
SVM with polynomial kernel and PCA | ADNI | 81.48% | [ |
SVM with RBF kernel | ADNI | 87% | [ |
SVM | ADNI | 74.67% for 2D and 78.67% for 3D | [ |
The dataset used in this paper comprises 160 structural MR scans accessed from the ADNI. All the images were T1-Weighted MPRAGE images belonging to the ADNI phase 1. Each image was downloaded in NIFTI format and contains images of AD, MCI, and CN. The data is shared
All the MRI scans were processed using Brain Suite Software. The motive of the pre-processing is to spatially normalize the brain into template space and remove unwanted brain parts. The steps involved in the pre-processing are as follows-
Radiomics is an approach to extracting a huge number of features from medical images with the help of data characterization algorithms. The Py-Radiomics [
Then,
Si represents the sum of the gaps between gray level i
This paper used Gaussian Naïve Bayes, K-Nearest neighbour’s, Support Vector Machine, XGBoost and Random Forest for the prediction of Alzheimer’s with Radiomic features.
It is a sub type of Naïve Bayes theorems. The term Naïve Bayes refers to a class of machine learning techniques that are built on the Bayes theorem. It uses Gaussian normal distribution as the probability distribution function.
It works on the basis of similarity of the features. It assumes that the similar objects are present closer. It computes the distance between the selected item and its neighbours and classifies them based on the computed distance [
SVM is used for prediction and regression purposes. But in general it has found more use in the classification purposes. SVM tries to find a hyperplane where it can separate different kinds of data by creating boundaries between them [
eXtreme Gradient Boosting employs a technique known as boosting to generate effective models. Boosting is an ensemble learning strategy that involves generating multiple weaker and simpler models in a row, with each new model trying to fix problems in the earlier model [
The random forest technique deploys ensemble learning that uses large decision tree-based classifiers to fix complex tasks. It is a collection of many decision trees based classifiers. The bagging technique is used for training the forest generated by the RF classifier [
Here, we proposed a random forest-based method for the early prediction of Alzheimer’s disease in elderly people. The Performance Measures of random forest are then compared with performance measures of Gaussian Naïve Bayes, K-NN, SVM, and XGBoost. The proposed random forest-based approach with the Radiomic features for AD
Performance | GNB | KNN | SVM | RF | XGBoost |
---|---|---|---|---|---|
Accuracy | 66% | 76% | 77% | 88% | 87% |
Precision | 68% | 77% | 77% | 88% | 87% |
Recall | 66% | 76% | 77% | 88% | 87% |
F1-score | 63% | 75% | 76% | 87% | 86% |
The proposed random forest based approach with the radiomic features for AD
Performance | GNB | KNN | SVM | RF | XGBoost |
---|---|---|---|---|---|
Accuracy | 57% | 65% | 69% | 72% | 64% |
Precision | 58% | 65% | 70% | 73% | 64% |
Recall | 57% | 65% | 69% | 72% | 64% |
F1-score | 49% | 65% | 68% | 71% | 63% |
The proposed random forest based approach with the radiomic features for MCI
Performance | GNB | KNN | SVM | RF | XGBoost |
---|---|---|---|---|---|
Accuracy | 66% | 60% | 64% | 69% | 65% |
Precision | 70% | 60% | 64% | 69% | 65% |
Recall | 66% | 60% | 64% | 69% | 65% |
F1-score | 63% | 60% | 64% | 69% | 65% |
In this paper, we calculated accuracy values for various numbers of trees (n_estimators) and it is observed that the best overall accuracy of 85% is obtained with a n_estimator value of 40, see
We calculated recall values for different numbers of trees (n_estimators) and it is observed that the best overall recall of 85% is obtained with a n_estimator value of 40, see
The proposed random forest-based approach with the Radiomic features achieved 85% accuracy recall, precision, and F1 score see
Performance | GNB | KNN | SVM | RF | XGBoost |
---|---|---|---|---|---|
Accuracy | 48% | 79% | 75% | 85% | 75% |
Precision | 77% | 81% | 77% | 85% | 73% |
Recall | 48% | 79% | 75% | 85% | 75% |
F1-score | 54% | 77% | 76% | 85% | 74% |
In the paper [
Reference | Dataset | method | Performance |
---|---|---|---|
[ |
ADNI | SVM with PCA | 80.9% |
[ |
ADNI | RNN | 81% |
[ |
ADNI | Logistic Regression | 79% |
Proposed method | ADNI | Random Forest | 85% |
Alzheimer’s disease is a frequently occurring mental illness that occurs in about 60–80%, cases of dementia. Depending upon the severity of symptoms the patients can be categorized in CN, MCI, and AD. Machine learning algorithms that are used in this paper are SVM taking parameter values as kernel = ‘linear’, random_state = 42, K- Nearest Neighbour taking parameter values as n_neighbors = 5, metric = ‘minkowski’, p = 2, Gaussian Naïve Bayes, XGBoost as well as Random Forest Classifier. Random Forest with parameter n_estimators = 40, criterion = ‘entropy’, random_state = 0, provides best result in terms of Accuracy. The proposed random forest-based approach with the Radiomic features achieved 85% accuracy. The proposed approach achieved 88% accuracy, 88% recall, 88% precision, and 87% F1-score for AD