Contributions
Abstract: P424
Type: Poster Sessions
Abstract Category: Pathology and pathogenesis of MS - Experimental models
Introduction: Disability can be prevented by early detection of multiple sclerosis lesions. Computer aided MS lesion detection via brain MRI segmentation is more efficient than time consuming manual methods. Yet, automatic segmentation is considered a challenging task due to the complexity in structures and similarity of normal tissues and lesions.
Methods: This paper proposes an automatic method of segmentation and active lesion detection for MS MRI images. The main method consists of three steps which includes image normalization and enhancement, region growing segmentation with gaussian contour enhancement and Hough transform. First, normalization of intensity is performed which is essential for the quantitative texture analysis of images. Then histogram normalization is implemented, which has better results compared to other methods. Later, region growing segmentation with gaussian contour enhancement is applied to segment white matter. It is a process which makes clusters of image regions according to image intensity of seed points. After selecting the initial seed, a statistical approach is used to grow the region. After this step, edges are detected by using the canny edge detector which helps us to apply Hough transformation. Edge points are taken as centre with a radius and then a circle is drawn across the lesion. The proposed method was evaluated on Gd-enhanced MRI data, taken from John Hunter Hospital of four patients with active MS. A total of 1973 MRI slices were used with each patient's scan yielding 665 to 800 slices. The auto-detection of lesions were compared to manual results from an expert radiologist.
Results: For evaluation, our algorithm consisted of four classes which are true negative, false negative, false positive and true positive, then overlap metric, structure similarity index (SSI) and precision were measured quantitatively. Overlap metric, SSI and precision were quantified as 0.91, 0.96 and 0.94, respectively. After auto-segmentation, active lesions were detected with an accuracy of 0.99 for examined images.
Conclusion: The proposed locally developed algorithm was able to and efficiently perform segmentation of white matter and accurately detect active MS lesion with high accuracy. This robust method will help radiologist to segment and detect active lesions automatically.
Disclosure: nothing to disclose
Abstract: P424
Type: Poster Sessions
Abstract Category: Pathology and pathogenesis of MS - Experimental models
Introduction: Disability can be prevented by early detection of multiple sclerosis lesions. Computer aided MS lesion detection via brain MRI segmentation is more efficient than time consuming manual methods. Yet, automatic segmentation is considered a challenging task due to the complexity in structures and similarity of normal tissues and lesions.
Methods: This paper proposes an automatic method of segmentation and active lesion detection for MS MRI images. The main method consists of three steps which includes image normalization and enhancement, region growing segmentation with gaussian contour enhancement and Hough transform. First, normalization of intensity is performed which is essential for the quantitative texture analysis of images. Then histogram normalization is implemented, which has better results compared to other methods. Later, region growing segmentation with gaussian contour enhancement is applied to segment white matter. It is a process which makes clusters of image regions according to image intensity of seed points. After selecting the initial seed, a statistical approach is used to grow the region. After this step, edges are detected by using the canny edge detector which helps us to apply Hough transformation. Edge points are taken as centre with a radius and then a circle is drawn across the lesion. The proposed method was evaluated on Gd-enhanced MRI data, taken from John Hunter Hospital of four patients with active MS. A total of 1973 MRI slices were used with each patient's scan yielding 665 to 800 slices. The auto-detection of lesions were compared to manual results from an expert radiologist.
Results: For evaluation, our algorithm consisted of four classes which are true negative, false negative, false positive and true positive, then overlap metric, structure similarity index (SSI) and precision were measured quantitatively. Overlap metric, SSI and precision were quantified as 0.91, 0.96 and 0.94, respectively. After auto-segmentation, active lesions were detected with an accuracy of 0.99 for examined images.
Conclusion: The proposed locally developed algorithm was able to and efficiently perform segmentation of white matter and accurately detect active MS lesion with high accuracy. This robust method will help radiologist to segment and detect active lesions automatically.
Disclosure: nothing to disclose