Contributions
Abstract: EP1352
Type: ePoster
Abstract Category: Clinical aspects of MS - MS Variants
Introduction: The study has been done to develop a new method for determining MS lesions using automatic segmentation of MR images.
Aim: TIA, T2A and FLAIR sequences of 100 MRI sections of female and male subjects randomly selected from GATA Haydarpasa Hospital database were used. Right now there is no automatic or semi-automatic extraction program used for determining MS lesions (large or small areas). Some MS plaques are quite difficult to detect. In this study, the detection and counting of MS lesions that are easily visible and those that are difficult to detect was done using this new method.
Method: By following multi-stage image processing steps, the extraction and counting of the MS lesions from the MR images was targeted. The background noise of the MRI scans,
Cleaning the background noise of the MRI scans, separation of the brain tissue from bone using skull mask, the removal of image defects on regions of interest (ROI) brain tissue, application of pixel-based classifier, “Otsu” threshold-based segmentation, histogram analysis and calculation of normal tissue and MS lesion sites, removing the region after region-based thresholding with MS lesions, and finally lesion counting using the abovementioned methods was done in each section.
Result: The success/performance of the segmentation was tested by ROC analysis. / performance tested. Two blinded radiologists marked lesion locations on raw data and were compared with proposed raw program outcomes on tissue. The areas marked by the radiologists were accepted as the correct localizations for MS lesions. The automatic segmentation algorithm that has been developed was put to ROC analysis for sensitivity and accuracy of location.
Disclosure: There is no conflict of interest
Abstract: EP1352
Type: ePoster
Abstract Category: Clinical aspects of MS - MS Variants
Introduction: The study has been done to develop a new method for determining MS lesions using automatic segmentation of MR images.
Aim: TIA, T2A and FLAIR sequences of 100 MRI sections of female and male subjects randomly selected from GATA Haydarpasa Hospital database were used. Right now there is no automatic or semi-automatic extraction program used for determining MS lesions (large or small areas). Some MS plaques are quite difficult to detect. In this study, the detection and counting of MS lesions that are easily visible and those that are difficult to detect was done using this new method.
Method: By following multi-stage image processing steps, the extraction and counting of the MS lesions from the MR images was targeted. The background noise of the MRI scans,
Cleaning the background noise of the MRI scans, separation of the brain tissue from bone using skull mask, the removal of image defects on regions of interest (ROI) brain tissue, application of pixel-based classifier, “Otsu” threshold-based segmentation, histogram analysis and calculation of normal tissue and MS lesion sites, removing the region after region-based thresholding with MS lesions, and finally lesion counting using the abovementioned methods was done in each section.
Result: The success/performance of the segmentation was tested by ROC analysis. / performance tested. Two blinded radiologists marked lesion locations on raw data and were compared with proposed raw program outcomes on tissue. The areas marked by the radiologists were accepted as the correct localizations for MS lesions. The automatic segmentation algorithm that has been developed was put to ROC analysis for sensitivity and accuracy of location.
Disclosure: There is no conflict of interest