Analysis of Textural Variations in Cerebellum in Brain to Identify Alzheimers by using Haralicks in Comparison with Gray Level Co-occurrence Matrix (GLRLM)

The aim of this research is to identify the textural variation in cerebellum of the brain to find the presence of Alzheimer's disease using Haralicks texture features in comparison with GLRLM texture features. For this analysis, MRI image dataset were extracted from OASIS database which consists of normal and Alzheimer's disease MRI images with sample size 50. The image dataset was used for feature extraction of texture images. Novel texture features are produced by the Haralicks and further extracted features are classified by KNN, SVM, Random Forest, Logistic Regression classifiers. From the results, novel texture features obtained from Haralicks provide best feature extraction from texture images such as mean values of normal is (0.62) and (0.54) for Alzheimer's. Loss in textural information is observed in the cerebellum of the brain. Classification using KNN classifiers, SVM classifiers, Random Forest classifier, Logistic Regression classifier for Haralicks features with accuracy (96%), Area Under Curve (AUC) (96%), F1-score (96%), precision (95%), recall (95%). The significance value is p<0.05. The G power is taken as 0.8. In this process we found that novel texture features extracted using Haralicks have performed better than the Gray Level Co-occurrence Matrix (GLRLM) texture features to identify the presence of Alzheimer's in the cerebellum of the brain MRI image.