Automatic brain extraction from MRI of human head scans using Helmholtz free energy principle and morphological operations. (February 2021)

• Record Type: Journal Article
• Authors: Ezhilarasan, K.; Praveenkumar, S.; Somasundaram, K.; Kalaiselvi, T.; Magesh, S.; Kiruthika, S.; Jeevarekha, A.
• Is Part Of: Biomedical signal processing and control. Volume 64(2021)
• Publication Date: 2021-02
• Subjects: Human brain -- Magnetic resonance imaging -- Helmholtz free energy -- Intensity transformation -- MRI -- Brain extraction -- Skull stripping; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2020.102270
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: An analogy between thermodynamic concepts for particles in a physical system and pixels in a still image is made. Based on Helmholtz free energy principle an intensity transform equation is made and is used to detect the brain boundary in an MRI and extract the brain. One of the proposed methods is able to extract the brain portion occurring in multiple regions of an MRI slice which is a challenging task for many brain extraction methods. 38 T1-W MRI datasets available in Internet Brain Segmentation Repository (IBSR) are used for experiments. The performance of the methods are evaluated by computing the Jaccard similarity index (J) and Dice coefficient (D). One of the proposed methods gave better or competitive results than that of the popular methods Brain Extraction Tool (BET) and Brain Surface Extractor (BSE) and few other recent methods. Both methods are robust against pepper and salt noise. Abstract: In this article, two novel methods are proposed to detect the brain boundary in a magnetic resonance image (MRI) by making an analogy between Helmholtz free energy (HFE) principle in thermodynamics and the intensity of pixels in an image. In the first method, an intensity transformation equation is derived using HFE. This transformation is applied to pixels in each slice of the MRI of human head and is used to detect boundaries of all objects present in it. To identify the brain portion among these objects, a sequence of two dimensional (2D) morphological operations and the largest connected component (LCC) analysis are done. These operations produced a mask for the brain using which the brain is extracted from the T1-W MRI slices. However, the LCC process failed in some cases to identify the brain region correctly. To address this issue, a second method which makes use of information in the adjacent slices is developed. This extended scheme performed well in extracting the brain portion from each slice of a MRI volume. Experiments are conducted by applying this schemes on 38 volumes of MRI of human head scans collected from the Internet Brain Segmentation Repository (IBSR). The proposed methods produced competitive or better results than the existing popular methods Brain Extraction Tool (BET), Brain Surface Extractor (BSE) and other similar recent methods.