Advanced methodology for multiple sclerosis lesion exploring: Towards a computer aided diagnosis system. (March 2019)

• Record Type: Journal Article
• Authors: Ghribi, Olfa; Maalej, Amira; Sellami, Lamia; Ben Slima, Mohamed; Maalej, Mohamed Ayman; Ben Mahfoudh, Khaireddine; Dammak, Mariem; Mhiri, Chokri; Ben Hamida, Ahmed
• Is Part Of: Biomedical signal processing and control. Volume 49(2019)
• Page Start: 274
• Page End: 288
• Publication Date: 2019-03
• Subjects: MRI magnetic resonance imagery -- MS multiple sclerosis -- GM gray matter -- WM white matter -- CSF cerebrospinal fluid -- FLAIR fluid attenuated inversion recovery image -- T1w T1-weighted image -- T2w T2-weighted image -- NABT normal appearing brain tissues -- GMM Gaussian mixture model -- UHST University Hospital Habib-Bourguiba in Sfax-Tunisia -- UNC University of North Carolina -- CHB Children's Hospital Boston -- TLL total lesion load (ml); Brain segmentation -- Lesion exploration -- Multiple sclerosis -- Gaussian mixture modal -- Lesion classification; 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.2018.12.010
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Automatic brain segmentation and MS lesion exploration. Advanced lesion parts discrimination into nucleus and edema. An automatic lesion classification according their lobar location. Various heterogeneous databases for evaluation. It compares positively with other publicly accessible MS lesion segmentation methods. Abstract: The exploration and diagnosis of severe neuropathologies is one of the essential tasks that could be reinforced and aided by advanced techniques in medical imagery. Cortex lesions and tumors are examples of pathologies where Magnetic Resonance Imagery modalities are the essential tools for exploration and tracking. Our main added value in this research field was focused on Multiple Sclerosis lesions exploration aiming to offer a clinical computer aided tool capable of helping clinicians during their daily explorations. An automatic method for brain segmentation and Multiple Sclerosis lesion exploration was in fact attentively conceived and tested. Firstly, the brain segmentation was performed using an atlas based Gaussian mixture model, followed by the calculation of a probabilistic lesion map on which an empirical threshold, characterized by a set of constraints, was applied together with a lesion refinement algorithm in order to create a binary lesion mask. The approach was applied to four clinical databases. Compared to various existing methods, the approach showed its efficiency in differentiating normal brain tissues with averages of 0.78 ± 0.04, 0.85 ± 0.04 and 0.89 ± 0.04 for Dice, sensitivity and specificity, respectively. Besides, it proved the possibility of accurately identifying lesions even in noisy images, and low lesion load, giving averages of 0.75 ± 0.04, 0.90 ± 0.08 and 0.75 ± 0 . 12 for the previous metrics, respectively. Furthermore, it is able to distinguish between the lesion center and the contiguous dirty white matter, which was carefully validated by clinicians. As a perspective, our approach could be extended to explore other neuropathologies such as Alzheimer's or Parkinson's diseases.