Contributions
Abstract: P483
Type: Poster
Abstract Category: Pathology and pathogenesis of MS - Imaging
Goal: This study is to assess the performance of FLAIR vs. FLAIR2 magnetic resonance (MR) images in their ability to identify lesions in multiple sclerosis (MS) patients using an auto-segmentation technique.
Methods: MR images, including 2D FLAIR and 2D T2w, from a randomly selected group of 50 relapsing-remitting MS patients were included. FLAIR2 images were created by multiplying co-registered FLAIR and T2 images using fslmaths. Lesion volumes for FLAIR and FLAIR2 were obtained using LesionTOADs. Lesions identified by a neuro-radiologist were used as the Gold Standard (GS). Performance indices relative to the GS, including DICE coefficient (reflecting both lesion location accuracy and size accuracy), Sensitivity (SEN) (reflecting the ability to identify GS lesions), Relative Lesion Load Volume (VOL) (relative volume difference to GS), and Symmetric Surface Distance (SSD) (distance between the GS lesions and auto-segmented lesions), were calculated for FLAIR and FLAIR2. Mean (meandiff) and corresponding 95% confidence interval (CI) were obtained for (FLAIR2 - FLAIR) differences. The Wilcoxson signed-rank test was used to statistically assess the differences. Relative improvement score (meandiff/meanFLAIR) for each index was calculated.
Results: Lesion loads could not be obtained for 3 out of the 50 patients due to running errors by LesionTOADs; these were excluded in subsequent calculations. Characteristics of the Gold Standard lesion volumes include mean=11,150 voxels, median=8,149 voxels, range: 139-48,706 voxels. SDD for one patient was undefined since no volume was identified by FLAIR. Overall, FLAIR2 had statistically significant and higher scores than FLAIR in all 4 indices; DICE: meandiff = 0.050 (95%CI: 0.013-0.087;
p value = 0.022) with a relative improvement of 14%; SEN: meandiff = 0.069 (95%CI: 0.040-0.098; p value < 0.001) with a relative improvement of 25%; VOL meandiff = -0.202 (95%CI: -0.352 - -0.049;
p value < 0.001) with a relative improvement of 31%; SSD: meandiff = -3.43mm (95%CI: -6.66mm - -0.20mm; p value = 0.018) with a relative improvement of 23%.
Conclusion: FLAIR2 outperforms FLAIR in its ability to provide lesion segmentation accuracy in MS patients. FLAIR2 can be easily calculated when T2 and FLAIR are available. FLAIR2 is a very simple way to increase overall segmentation performance with little additional effort when using LesionTOADs.
Disclosure:
Megan Le: nothing to disclose
Alexander Rauscher: nothing to disclose
Tom Brosch: nothing to disclose
Youngjin Yoo: nothing to disclose
Lisa Tang: nothing to disclose
Mike Jarrett: nothing to disclose
Anthony Traboulsee has received personal compensation (honorarium) consulting for Genzyme, Roche, Teva, and Biogen and research support (PI on clinical trials) from Genzyme, Roche, Chugai, and Biogen
David Li has received research funding from the Canadian Institute of Health Research and Multiple Sclerosis Society of Canada. He is the Director of the UBC MS/MRI Research Group which has been contracted to perform central analysis of MRI scans for therapeutic trials with Novartis, Perceptives, Roche and Sanofi-Aventis. The UBC MS/MRI Research Group has also received grant support for investigator-initiated independent studies from Genzyme, Merck-Serono, Novartis and Roche. He has acted as a consultant to Vertex Pharmaceuticals and served on the Data and Safety Advisory Board for Opexa Therapeutics and Scientific Advisory Boards for Adelphi Group, Novartis and Roche. He has also given lectures which have been supported by non-restricted education grants from Novartis and Biogen.
Roger Tam: nothing to disclose
Abstract: P483
Type: Poster
Abstract Category: Pathology and pathogenesis of MS - Imaging
Goal: This study is to assess the performance of FLAIR vs. FLAIR2 magnetic resonance (MR) images in their ability to identify lesions in multiple sclerosis (MS) patients using an auto-segmentation technique.
Methods: MR images, including 2D FLAIR and 2D T2w, from a randomly selected group of 50 relapsing-remitting MS patients were included. FLAIR2 images were created by multiplying co-registered FLAIR and T2 images using fslmaths. Lesion volumes for FLAIR and FLAIR2 were obtained using LesionTOADs. Lesions identified by a neuro-radiologist were used as the Gold Standard (GS). Performance indices relative to the GS, including DICE coefficient (reflecting both lesion location accuracy and size accuracy), Sensitivity (SEN) (reflecting the ability to identify GS lesions), Relative Lesion Load Volume (VOL) (relative volume difference to GS), and Symmetric Surface Distance (SSD) (distance between the GS lesions and auto-segmented lesions), were calculated for FLAIR and FLAIR2. Mean (meandiff) and corresponding 95% confidence interval (CI) were obtained for (FLAIR2 - FLAIR) differences. The Wilcoxson signed-rank test was used to statistically assess the differences. Relative improvement score (meandiff/meanFLAIR) for each index was calculated.
Results: Lesion loads could not be obtained for 3 out of the 50 patients due to running errors by LesionTOADs; these were excluded in subsequent calculations. Characteristics of the Gold Standard lesion volumes include mean=11,150 voxels, median=8,149 voxels, range: 139-48,706 voxels. SDD for one patient was undefined since no volume was identified by FLAIR. Overall, FLAIR2 had statistically significant and higher scores than FLAIR in all 4 indices; DICE: meandiff = 0.050 (95%CI: 0.013-0.087;
p value = 0.022) with a relative improvement of 14%; SEN: meandiff = 0.069 (95%CI: 0.040-0.098; p value < 0.001) with a relative improvement of 25%; VOL meandiff = -0.202 (95%CI: -0.352 - -0.049;
p value < 0.001) with a relative improvement of 31%; SSD: meandiff = -3.43mm (95%CI: -6.66mm - -0.20mm; p value = 0.018) with a relative improvement of 23%.
Conclusion: FLAIR2 outperforms FLAIR in its ability to provide lesion segmentation accuracy in MS patients. FLAIR2 can be easily calculated when T2 and FLAIR are available. FLAIR2 is a very simple way to increase overall segmentation performance with little additional effort when using LesionTOADs.
Disclosure:
Megan Le: nothing to disclose
Alexander Rauscher: nothing to disclose
Tom Brosch: nothing to disclose
Youngjin Yoo: nothing to disclose
Lisa Tang: nothing to disclose
Mike Jarrett: nothing to disclose
Anthony Traboulsee has received personal compensation (honorarium) consulting for Genzyme, Roche, Teva, and Biogen and research support (PI on clinical trials) from Genzyme, Roche, Chugai, and Biogen
David Li has received research funding from the Canadian Institute of Health Research and Multiple Sclerosis Society of Canada. He is the Director of the UBC MS/MRI Research Group which has been contracted to perform central analysis of MRI scans for therapeutic trials with Novartis, Perceptives, Roche and Sanofi-Aventis. The UBC MS/MRI Research Group has also received grant support for investigator-initiated independent studies from Genzyme, Merck-Serono, Novartis and Roche. He has acted as a consultant to Vertex Pharmaceuticals and served on the Data and Safety Advisory Board for Opexa Therapeutics and Scientific Advisory Boards for Adelphi Group, Novartis and Roche. He has also given lectures which have been supported by non-restricted education grants from Novartis and Biogen.
Roger Tam: nothing to disclose