- Email: [email protected]
Tracking white matter degeneration in frontotemporal dementia using serial diffusion tensor imaging
P56
Alzheimer’s Imaging Consortium Poster Presentations: IC-P
Figure 1. (B): Visualization of thickness distribution of the three groups in the most significantly different (least p-value) between each pair (for each each row). It is obvious to see there is large overlap, which is the primary reason for moderate classification performance.
optimize the cost-ratio, C and kernel width, was trained on these features. We assigned a higher cost for false negatives compared to false positives. We performed 2-fold stratified cross-validation, repeated 100 times, to evaluate performance. Results: CTHK partition features showed significant overlap among the three groups (Figure 1,B). Classification based on these features achieved an accuracy of 50% to 60% at best (Figure 1,C and Figure 1D). Conclusions: We present the first study on assessing the classification power of CTHK for the sub-classification of aMCI. Although previous studies have shown the presence of group differences, the results of this study show that the differences are quite subtle. In fact, our results using a classifier based on the powerful cost-sensitive SVM show that discrimination between single and multiple domain aMCI is limited using CTHK measures. References: 1.B€ackman, L., Jones, S., Berger, A.-K., Laukka, E. J., & Small, B. J. (2004). MulAple cogniAve deficits during the transiAon to Alzheimer’s disease. Journal of Internal Medicine, 256(3), 195–204. http://dx.doi.org/10.1111/j. 1365-2796.2004.01386.x. 2. Bell-McGinty, S., Lopez, O. L., Meltzer, C. C., Scanlon, J. M., Whyte, E. M., Dekosky, S. T., & Becker, J. T. (2005). DifferenAal corAcal atrophy in subgroups of mild cogniAve impairment. Archives of Neurology, 62(9), 1393–1397. http://dx.doi.org/10.1001/archneur.62.9.1393. 3. FennemaNotesAne, C., Hagler, D. J., Mcevoy, L. K., Fleisher, A. S., Wu, E. H., Karow, D. S., & Dale, A. M. (2009). Structural MRI biomarkers for preclinical and mild Alzheimer’s disease. Human Brain Mapping, 30(10), 3238–3253. http://dx.doi.org/10.1002/hbm.20744. 4. Seo, S., Im, K., Lee, J., Kim, Y., Kim, S., Kim, S., et al. (2007). CorAcal thickness in single-versus mulAple-domain amnesAc mild cogniAve impairment. NeuroImage, 36(2), 289–297. 5. Whitwell, J., Petersen, R., & Negash, S. (2007). Paherns of atrophy differ among specific subtypes of mild cogniAve impairment. Archives of Neurology, 1–9. 6. Gibson E, Wang L, Beg MF. CorAcal thickness measurement using Eulerian PDEs and surfacebased global topological informa Aon. Org Human Brain Mapping, 15th Ann MeeAng. 2009. 7. Sachdev PS, Brodaty H, Reppermund S, Kochan NA, Trollor JN, Draper B, et al. The Sydney Memory and Ageing Study (MAS): methodology and baseline medical and neuropsychiatric characterisAcs of an elderly epidemiological non-demented cohort of Australians aged 70–90 years. InternaAonal Psychogeriatrics. 2010 Dec;22(08):1248–1264. 8. Fitzpatrick M, Sonka M. Handbook of medical imaging vol 2: Medical image processing & analysis (PM80). SPIEInternaAonal Society for OpAcal Engineering; 2000.
IC-P-098
Figure 1. (C): ROC comparison with best performance for each pair ranked by AUC. The performance for NC vs. sd-aMCI and sd- vs. md-aMCI is just about random and only the classifier NC vs. md-aMCI performs better than random. This is expected due to large overlap in thickness features found among the three groups, as shown in Figure 1(B).
Figure 1. (D): Comparison of best classification performance for each pair, in terms of various metrics.
TRACKING WHITE MATTER DEGENERATION IN FRONTOTEMPORAL DEMENTIA USING SERIAL DIFFUSION TENSOR IMAGING
Colin Mahoney1, Hui Zhang2, Ian Malone1, Jennifer Nicholas2, Nicole Schmitz3, Jonathan Schott4, Jason Warren2, Nick Fox1, 1UCL Institute of Neurology, London, United Kingdom; 2University College London, London, United Kingdom; 3Dementia Research Centre, London, United Kingdom; 4Dementia Research Centre, Institute of Neurology, UCL, London, United Kingdom. Contact e-mail: [email protected] Background: Identification of novel imaging biomarkers for monitoring progression in neurodegenerative disease is an important research focus. Diffusion tensor imaging (DTI) allows measurement of white matter (WM) microstructure and is a potential outcome measure in future treatment trials. Longitudinal DTI studies in Frontotemporal Dementia (FTD) are lacking. Here we report changes in WM microstructure in a cohort with FTD using serial DTI. Methods: 15 individuals with FTD (mean age¼61.7, standard deviation (SD) 11.2; n¼5 MAPT mutations; n¼4 C9ORF72 mutations) underwent two DTI scans (interval¼1.3yrs SD¼0.4) and were compared with 15 healthy individuals (mean age¼62.9yrs, SD¼8.2; interval¼1.4yrs, SD¼0.4). A within-subject mean tensor volume was created from baseline and follow-up images using DTITK: a state-of-the-art unbiased longitudinal analysis framework for DTI. A group-wise atlas was created from each subject’s mean tensor volume.
Alzheimer’s Imaging Consortium Poster Presentations: IC-P Baseline and follow-up tensor volumes were then registered to the groupwise template. A fractional anisotropy (FA) map was created for each registered tensor volume. Regions of interest were assessed by registering the ICBM-DTI-81 atlas to the final template FA map. Annualised rates of change were calculated and assessed in a linear regression model. Results: A significant reduction (p¼0.03) in FA of 1.3%/year (SD¼2.2) was observed across global WM in those with FTD (controls -0.4%/year (SD¼2.1)). Within WM tracts, across the whole FTD group, the most statistically significant reduction (p¼0.006) was 2.9%/year (SD¼4.1) within the left inferior longitudinal fasciculus (ILF) (controls -1.1%/year (SD¼1.0)). In those with a MAPT mutation, the most significant reduction (p¼0.04) was within the left uncinate fasciculus (UF) (6.5%/year (SD¼5.1); controls¼-0.5%/year (SD¼4.0)). In those with a C9ORF72 mutation, the most significant reduction was within the superior cerebellar peduncles (SCP) (right 9.4%/year (SD¼5.5), controls¼0.3%/year (64.2), p¼0.02; left 9.5%/year (SD¼9.6) controls¼0.4%/year (SD¼4.5), p¼0.06). Conclusions: This study demonstrates for the first time the utility of DTI for tracking progression of regional WM tract change in FTD. Marked differences emerged between groups. Across FTD the ILF emerged as a commonly affected tract. Within genetically defined groups, particular tracts emerged: the UF in MAPT and the SCP in C9ORF72. Longitudinal DTI can detect microstructural change in WM and may be a useful biomarker in treatment trials. IC-P-099
TRAINING FOR MANUAL HIPPOCAMPAL SEGMENTATION BASED ON THE EADC-ADNI HARMONIZED PROTOCOL
Marina Boccardi1, Nicolas Robitaille2, Fernando Valdivia3, Martina Bocchetta4, Corinna Bauer5, Melanie Blair6, Emma Burton7, Enrica Cavedo4, Adam Christensen8, Kristian Steen Frederiksen9, Michel Grothe10, Mariangela Lanfredi11, Yawu Liu12, Oliver Martinez13, Masami Nishikawa14, Marileen Portegies15, Margo Pronk16, Travis Stoub17, Tim Swihart18, Chad Ward19, Liana Apostolova20, Rossana Ganzola21, Gregory Preboske19, Dominik Wolf22, Clifford Jack19, Simon Duchesne2, Giovanni Frisoni23, 1IRCCS S.Giovanni di Dio-Fatebenefratelli, Brescia, Italy; 2Universite Laval and Centre de Recherche Universite Laval-Robert Giffard, Quebec, Quebec, Canada; 3Department of Radiology, Universite Laval and Centre de Recherche de l’Institut Universitaire de Sante Mentale de Quebec, Quebec City, Quebec, Canada; 4IRCCS Fatebenefratelli, Brescia, Italy; 5Boston Univeristy School of Medicine, Boston, Massachusetts, United States; 6University of Exeter, Exeter, United Kingdom; 7Newcastle University, Newcastle Upon Tyne, United Kingdom; 8 Northwestern University, Chicago, Illinois, United States; 9Memory Disorders Research Group, Department of Neurology, Rigshospitalet, Copenhagen, Denmark; 10University of Rostock, Rostock, Germany; 11 IRCCS S.Giovanni di Dio-Fatebenefratelli, Brescia, Italy; 12Kuopio University Hospital, Kuopio, Finland; 13UC Davis, Davis, California, United States; 14Kawamura Gakuen Woman’s University, Abiko-City, Japan; 15University Medical Center, Utrecht, Netherlands; 16Image Analysis Centre-Vrije Universiteit Medisch Centrum, Amsterdam, Netherlands; 17Rush University Medical Center, Chicago, Illinois, United States; 18Layton Aging and Alzheimer’s Disease Center, Oregon Health and Science University, Portland, Oregon, United States; 19Mayo Clinic, Rochester, Minnesota, United States; 20UCLA, Los Angeles, California, United States; 21Universite Laval and Centre de Recherche Universite Laval-Robert Giffard, Quebec City, Quebec, Canada; 22University Medical Center Mainz, Mainz, Germany; 23IRCCS Fatebenefratelli, Brescia, Italy. Contact e-mail: [email protected] Background: A Harmonized Protocol (HP) for manual hippocampal segmentation was defined by a Delphi panel (Boccardi et al., Neurology 2012:78(S1):S04003) and described in a written document. Benchmark images were segmented by 5 HP-expert tracers as the reference gold standard and uploaded on a web-based platform providing a standard training system. Methods: Seventeen "Na€ıve" tracers with ICC>0.80 in hippocampal segmentation based on their local protocols and no previous experience with the HP were recruited from EADC/ADNI (European Alzheimer Disease
P57
Figure. Sample images of tracings and quantitative visual feedback vs benchmark segmentations (blue dotted lines). Red¼far from benchmark; green¼similar to benchmark; pink¼to be evaluated qualitatively. Consortium/Alzheimer’s Disease Neuroimaging Initiative) centres. They were provided with the same HP criteria and segmentation instructions through the web-platform, and with the same version of MultiTracer for manual segmentation.Training images from 10 ADNI subjects for whom benchmark labels were available were divided into three rounds (n¼2, n¼4, n¼4), balanced by magnetic strength field and degree of hippocampal atrophy. Tracers were asked to segment both hippocampi based on the HP and upload segmented images on the platform. Visual feedback was provided showing point by point discrepancies of segmentations versus the reference in color code.Written feedback was provided slice by slice in addition to the visual feedback for the first two rounds. In subsequent rounds tracers were asked to upload the images of the previous round that had been corrected based on the feedback, and to segment 4 additional images.Dice and Jaccard overlapping indices were computed versus the mean of the HP-experts’ segmentations. A slice by slice visual quality check (QC) was carried out to match overlapping values with levels of compliance with
Table Median, minimum and maximum Dice and Jaccard overlapping values and ICC inter-rater values (consistency method) in the third and last training round, completed by 10 tracers, computed with and without segmentations corrected from previous training rounds. Overlapping indices versus Benchmark segmentations
Inter-rater ICCs among the 10 “Na€ıve” tracers (95% CI)
3 Tesla
3 Tesla + 1.5 Tesla
1.5 Tesla
Dice Jaccard Dice Jaccard Right
left
N¼10 MRIs – 6 corrected segmentations from previous training rounds included Median 0.91 0.83 Minimum 0.89 0.80 Maximum 0.92 0.85
0.90 0.81 0.82 0.74 0.91 0.84
0.98 (0.95-0.99) 0.98(0.95-0.99)
N¼4 new MRIs – corrected segmentations from previous training rounds excluded Median 0.90 0.82 Minimum 0.88 0.78 Maximum 0.92 0.86
0.90 0.82 0.71 0.63 0.92 0.86
0.89 (0.69-0.99) 0.91 (0.72-0.99)
Alzheimer’s Imaging Consortium Poster Presentations: IC-P
Figure 1. (B): Visualization of thickness distribution of the three groups in the most significantly different (least p-value) between each pair (for each each row). It is obvious to see there is large overlap, which is the primary reason for moderate classification performance.
optimize the cost-ratio, C and kernel width, was trained on these features. We assigned a higher cost for false negatives compared to false positives. We performed 2-fold stratified cross-validation, repeated 100 times, to evaluate performance. Results: CTHK partition features showed significant overlap among the three groups (Figure 1,B). Classification based on these features achieved an accuracy of 50% to 60% at best (Figure 1,C and Figure 1D). Conclusions: We present the first study on assessing the classification power of CTHK for the sub-classification of aMCI. Although previous studies have shown the presence of group differences, the results of this study show that the differences are quite subtle. In fact, our results using a classifier based on the powerful cost-sensitive SVM show that discrimination between single and multiple domain aMCI is limited using CTHK measures. References: 1.B€ackman, L., Jones, S., Berger, A.-K., Laukka, E. J., & Small, B. J. (2004). MulAple cogniAve deficits during the transiAon to Alzheimer’s disease. Journal of Internal Medicine, 256(3), 195–204. http://dx.doi.org/10.1111/j. 1365-2796.2004.01386.x. 2. Bell-McGinty, S., Lopez, O. L., Meltzer, C. C., Scanlon, J. M., Whyte, E. M., Dekosky, S. T., & Becker, J. T. (2005). DifferenAal corAcal atrophy in subgroups of mild cogniAve impairment. Archives of Neurology, 62(9), 1393–1397. http://dx.doi.org/10.1001/archneur.62.9.1393. 3. FennemaNotesAne, C., Hagler, D. J., Mcevoy, L. K., Fleisher, A. S., Wu, E. H., Karow, D. S., & Dale, A. M. (2009). Structural MRI biomarkers for preclinical and mild Alzheimer’s disease. Human Brain Mapping, 30(10), 3238–3253. http://dx.doi.org/10.1002/hbm.20744. 4. Seo, S., Im, K., Lee, J., Kim, Y., Kim, S., Kim, S., et al. (2007). CorAcal thickness in single-versus mulAple-domain amnesAc mild cogniAve impairment. NeuroImage, 36(2), 289–297. 5. Whitwell, J., Petersen, R., & Negash, S. (2007). Paherns of atrophy differ among specific subtypes of mild cogniAve impairment. Archives of Neurology, 1–9. 6. Gibson E, Wang L, Beg MF. CorAcal thickness measurement using Eulerian PDEs and surfacebased global topological informa Aon. Org Human Brain Mapping, 15th Ann MeeAng. 2009. 7. Sachdev PS, Brodaty H, Reppermund S, Kochan NA, Trollor JN, Draper B, et al. The Sydney Memory and Ageing Study (MAS): methodology and baseline medical and neuropsychiatric characterisAcs of an elderly epidemiological non-demented cohort of Australians aged 70–90 years. InternaAonal Psychogeriatrics. 2010 Dec;22(08):1248–1264. 8. Fitzpatrick M, Sonka M. Handbook of medical imaging vol 2: Medical image processing & analysis (PM80). SPIEInternaAonal Society for OpAcal Engineering; 2000.
IC-P-098
Figure 1. (C): ROC comparison with best performance for each pair ranked by AUC. The performance for NC vs. sd-aMCI and sd- vs. md-aMCI is just about random and only the classifier NC vs. md-aMCI performs better than random. This is expected due to large overlap in thickness features found among the three groups, as shown in Figure 1(B).
Figure 1. (D): Comparison of best classification performance for each pair, in terms of various metrics.
TRACKING WHITE MATTER DEGENERATION IN FRONTOTEMPORAL DEMENTIA USING SERIAL DIFFUSION TENSOR IMAGING
Colin Mahoney1, Hui Zhang2, Ian Malone1, Jennifer Nicholas2, Nicole Schmitz3, Jonathan Schott4, Jason Warren2, Nick Fox1, 1UCL Institute of Neurology, London, United Kingdom; 2University College London, London, United Kingdom; 3Dementia Research Centre, London, United Kingdom; 4Dementia Research Centre, Institute of Neurology, UCL, London, United Kingdom. Contact e-mail: [email protected] Background: Identification of novel imaging biomarkers for monitoring progression in neurodegenerative disease is an important research focus. Diffusion tensor imaging (DTI) allows measurement of white matter (WM) microstructure and is a potential outcome measure in future treatment trials. Longitudinal DTI studies in Frontotemporal Dementia (FTD) are lacking. Here we report changes in WM microstructure in a cohort with FTD using serial DTI. Methods: 15 individuals with FTD (mean age¼61.7, standard deviation (SD) 11.2; n¼5 MAPT mutations; n¼4 C9ORF72 mutations) underwent two DTI scans (interval¼1.3yrs SD¼0.4) and were compared with 15 healthy individuals (mean age¼62.9yrs, SD¼8.2; interval¼1.4yrs, SD¼0.4). A within-subject mean tensor volume was created from baseline and follow-up images using DTITK: a state-of-the-art unbiased longitudinal analysis framework for DTI. A group-wise atlas was created from each subject’s mean tensor volume.
Alzheimer’s Imaging Consortium Poster Presentations: IC-P Baseline and follow-up tensor volumes were then registered to the groupwise template. A fractional anisotropy (FA) map was created for each registered tensor volume. Regions of interest were assessed by registering the ICBM-DTI-81 atlas to the final template FA map. Annualised rates of change were calculated and assessed in a linear regression model. Results: A significant reduction (p¼0.03) in FA of 1.3%/year (SD¼2.2) was observed across global WM in those with FTD (controls -0.4%/year (SD¼2.1)). Within WM tracts, across the whole FTD group, the most statistically significant reduction (p¼0.006) was 2.9%/year (SD¼4.1) within the left inferior longitudinal fasciculus (ILF) (controls -1.1%/year (SD¼1.0)). In those with a MAPT mutation, the most significant reduction (p¼0.04) was within the left uncinate fasciculus (UF) (6.5%/year (SD¼5.1); controls¼-0.5%/year (SD¼4.0)). In those with a C9ORF72 mutation, the most significant reduction was within the superior cerebellar peduncles (SCP) (right 9.4%/year (SD¼5.5), controls¼0.3%/year (64.2), p¼0.02; left 9.5%/year (SD¼9.6) controls¼0.4%/year (SD¼4.5), p¼0.06). Conclusions: This study demonstrates for the first time the utility of DTI for tracking progression of regional WM tract change in FTD. Marked differences emerged between groups. Across FTD the ILF emerged as a commonly affected tract. Within genetically defined groups, particular tracts emerged: the UF in MAPT and the SCP in C9ORF72. Longitudinal DTI can detect microstructural change in WM and may be a useful biomarker in treatment trials. IC-P-099
TRAINING FOR MANUAL HIPPOCAMPAL SEGMENTATION BASED ON THE EADC-ADNI HARMONIZED PROTOCOL
Marina Boccardi1, Nicolas Robitaille2, Fernando Valdivia3, Martina Bocchetta4, Corinna Bauer5, Melanie Blair6, Emma Burton7, Enrica Cavedo4, Adam Christensen8, Kristian Steen Frederiksen9, Michel Grothe10, Mariangela Lanfredi11, Yawu Liu12, Oliver Martinez13, Masami Nishikawa14, Marileen Portegies15, Margo Pronk16, Travis Stoub17, Tim Swihart18, Chad Ward19, Liana Apostolova20, Rossana Ganzola21, Gregory Preboske19, Dominik Wolf22, Clifford Jack19, Simon Duchesne2, Giovanni Frisoni23, 1IRCCS S.Giovanni di Dio-Fatebenefratelli, Brescia, Italy; 2Universite Laval and Centre de Recherche Universite Laval-Robert Giffard, Quebec, Quebec, Canada; 3Department of Radiology, Universite Laval and Centre de Recherche de l’Institut Universitaire de Sante Mentale de Quebec, Quebec City, Quebec, Canada; 4IRCCS Fatebenefratelli, Brescia, Italy; 5Boston Univeristy School of Medicine, Boston, Massachusetts, United States; 6University of Exeter, Exeter, United Kingdom; 7Newcastle University, Newcastle Upon Tyne, United Kingdom; 8 Northwestern University, Chicago, Illinois, United States; 9Memory Disorders Research Group, Department of Neurology, Rigshospitalet, Copenhagen, Denmark; 10University of Rostock, Rostock, Germany; 11 IRCCS S.Giovanni di Dio-Fatebenefratelli, Brescia, Italy; 12Kuopio University Hospital, Kuopio, Finland; 13UC Davis, Davis, California, United States; 14Kawamura Gakuen Woman’s University, Abiko-City, Japan; 15University Medical Center, Utrecht, Netherlands; 16Image Analysis Centre-Vrije Universiteit Medisch Centrum, Amsterdam, Netherlands; 17Rush University Medical Center, Chicago, Illinois, United States; 18Layton Aging and Alzheimer’s Disease Center, Oregon Health and Science University, Portland, Oregon, United States; 19Mayo Clinic, Rochester, Minnesota, United States; 20UCLA, Los Angeles, California, United States; 21Universite Laval and Centre de Recherche Universite Laval-Robert Giffard, Quebec City, Quebec, Canada; 22University Medical Center Mainz, Mainz, Germany; 23IRCCS Fatebenefratelli, Brescia, Italy. Contact e-mail: [email protected] Background: A Harmonized Protocol (HP) for manual hippocampal segmentation was defined by a Delphi panel (Boccardi et al., Neurology 2012:78(S1):S04003) and described in a written document. Benchmark images were segmented by 5 HP-expert tracers as the reference gold standard and uploaded on a web-based platform providing a standard training system. Methods: Seventeen "Na€ıve" tracers with ICC>0.80 in hippocampal segmentation based on their local protocols and no previous experience with the HP were recruited from EADC/ADNI (European Alzheimer Disease
P57
Figure. Sample images of tracings and quantitative visual feedback vs benchmark segmentations (blue dotted lines). Red¼far from benchmark; green¼similar to benchmark; pink¼to be evaluated qualitatively. Consortium/Alzheimer’s Disease Neuroimaging Initiative) centres. They were provided with the same HP criteria and segmentation instructions through the web-platform, and with the same version of MultiTracer for manual segmentation.Training images from 10 ADNI subjects for whom benchmark labels were available were divided into three rounds (n¼2, n¼4, n¼4), balanced by magnetic strength field and degree of hippocampal atrophy. Tracers were asked to segment both hippocampi based on the HP and upload segmented images on the platform. Visual feedback was provided showing point by point discrepancies of segmentations versus the reference in color code.Written feedback was provided slice by slice in addition to the visual feedback for the first two rounds. In subsequent rounds tracers were asked to upload the images of the previous round that had been corrected based on the feedback, and to segment 4 additional images.Dice and Jaccard overlapping indices were computed versus the mean of the HP-experts’ segmentations. A slice by slice visual quality check (QC) was carried out to match overlapping values with levels of compliance with
Table Median, minimum and maximum Dice and Jaccard overlapping values and ICC inter-rater values (consistency method) in the third and last training round, completed by 10 tracers, computed with and without segmentations corrected from previous training rounds. Overlapping indices versus Benchmark segmentations
Inter-rater ICCs among the 10 “Na€ıve” tracers (95% CI)
3 Tesla
3 Tesla + 1.5 Tesla
1.5 Tesla
Dice Jaccard Dice Jaccard Right
left
N¼10 MRIs – 6 corrected segmentations from previous training rounds included Median 0.91 0.83 Minimum 0.89 0.80 Maximum 0.92 0.85
0.90 0.81 0.82 0.74 0.91 0.84
0.98 (0.95-0.99) 0.98(0.95-0.99)
N¼4 new MRIs – corrected segmentations from previous training rounds excluded Median 0.90 0.82 Minimum 0.88 0.78 Maximum 0.92 0.86
0.90 0.82 0.71 0.63 0.92 0.86
0.89 (0.69-0.99) 0.91 (0.72-0.99)