Evaluation of thresholding techniques on 3D fractal dimension MRI images

• Main Authors: Jamaludin, Iqbal, Che Azemin, Mohd Zulfaezal, Musa, Liyana, Sapuan, Abdul HaliM, Abdul Majid, Zafri Azran, Ahmad, Azmir, A. Rahman, Nor Azlina, Sideek, Mohamed Arshad
• Format: Article
• Language: English; English; English
• Published: Medwell Publications 2017
• Subjects: QA75 Electronic computers. Computer science; QM Human anatomy
• Online Access: http://irep.iium.edu.my/61420/1/2017%20iSHAMS.pdf; http://irep.iium.edu.my/61420/11/61420_Evaluation%20of%20thresholding%20techniques_scopus.pdf; http://irep.iium.edu.my/61420/16/61420_Evaluation%20of%20thresholding%20techniques%20on%203D_scopus.pdf; http://irep.iium.edu.my/61420/; http://docsdrive.com/pdfs/medwelljournals/jeasci/2017/6387-6392.pdf

Fractal Analysis (FA) application in medical field has been used to measure the occurrence of changes in the brain complexity for some diseased and normal aging brain. Within FA, there is Fractal Dimension (FD) which is an index of structural complexity. FD methods were classified into three major categories which are box-counting method, variance method and spectral method. Here in this study we used box-counting method to measure our FD due to its directness and automatic computability capabilities. We evaluated the impact of different thresholding techniques when quantifying the FD. We applied three different thresholding techniques on our brain MRI images: Otsu’s method, midpoint method and hysteresis method. A total of 27 subjects (14 males and 13 females) aged ranging between 21-25 years old were voluntarily participated. The process of thresholding the images and computation of the FD values were done under MATLAB. There was a statistically significant difference between groups of thresholding techniques as determined by Friedman test with χ(2) = 48.667 and p<0.001. Post-hoc Wilcoxon-signed ranked test with Bonferroni correction done where the p-value was set to 0.017. All three group pairs (midpoint vs. Otsu’s, midpoint vs. hysteresis and hysteresis vs. Otsu’s) were statistically significant different with p<0.001. Pearson’s correlation showed moderate correlation (r = 0.446, p = 0.02) between hysteresis and Otsu’s. Spearman correlation showed weak correlation which is not significant (r = 0.256, p = 0.198) for midpoint vs. Otsu’s and (r = 0.252, p = 0.204) for midpoint vs. hysteresis. In conclusion, different thresholding techniques do have impact on FD values but with moderate correlation between them.