_________________________________________________________
Supervisors: Elna-Marie Larsson, ProfessorSection of Neuroradiology Department of radiology Uppsala University Hospital Uppsala, Sweden
Master Thesis in Medicine
(Second Level, 15 Credits)
Intracranial volume Segmentation
2
Intracranial volume Segmentation
Keywords
:Magnetic Resonance MR, Intracranial volume ICV, Total intracranial volume TIV List of abbreviations used
Magnetic Resonance (MR)
front temporal lobar degeneration (FTLD)
Huntington's disease (HD)
Alzheimer's disease AD
Total Intracranial volume TIV
Proton density weighted Imaged (PD weighted MR images)
Intracranial volume (ICV)
mesial Temporal Lobe Epilepsy (mTLE)
Ventricle-Brain Ratio (VBR)
Cerebrospinal fluid (CSF)
Statistical Parametric Mapping (SPM)
Prospective Investigation of the Vasculature in Uppsala Seniors (PIVUS)
Time of Echo (TE)
Intraclass correlation coefficient (ICC)
3
Contents
Intracranial volume Segmentation ... 1
Intracranial volume Segmentation ... 2
Keywords ... 2
Abstract ... 4
Introduction ... 5
Material and methods ... 6
Subjects ... 6
Imaging ... 6
Image analysis ... 7
ICV was measured using SmartPaint, ... 7
Statistical analysis ... 8
Results ... 8
Discussion ... 10
Conclusion ... 11
4
Abstract
Background and Purpose MR-based volumetric measure of total intracranial volume (TIV) is increasingly used for normalization of brain structure volume in research to evaluate various brain diseases; such as brain atrophy, frontotemporal lobar degeneration (FTLD), Huntington's disease (HD) and Alzheimer's disease (AD). TIV is used to compensate for the variation in individual head size. Our purpose was to retrospectively evaluate the intracranial volume segmentation software SmartPaint.
Material and Methods Forty healthy subjects (mean age, 77.5 ±2.5 years) included in this study were examined with T1-weighted and dual echo sequences (proton density (PD) and T2) using 1.5 T magnetic resonance imaging (MR). TIV were measured by two operators using the semi-automatic segmentation tool Smart paint. Paired t-test was used to assess the difference between the two observers measurements. Subject’s data were acquired at age of 75 years and at age 80 years. Informed consent was obtained from all subjects and the study had local ethics committee approval.
Results There was no significant between two operators’ measures (P>0.08), difference in mean between the two segmented ICV was close to zero, and the coefficient of variance was less than 1%.
Conclusion Segmentation results indicate high correlation and accuracy between the two operators; therefore, the software can be used for segmenting ICV tissue.
5
Introduction
Quantitative volume analysis using software for measurements on magnetic resonance imaging (MRI) is increasingly used to diagnose brain atrophy and measure volumes of cerebral regions (1). Brain atrophy is a major characteristic feature of degenerative diseases and dementia such as Alzheimer's disease (AD) (2), Huntington's disease (HD) (3), and frontotemporal lobar degeneration (FTLD) (4). Cerebral atrophy is loss and degeneration of neurons and the network between them, and it could result in general brain shrinkage as in AD, or could be regional or focal as in hippocampal atrophy. Analysis of head size and brain structure faces two major challenges; the different head size among individuals according to gender, body and age, and the analysis software.(5)
Various types of brain segmentation are performed, such as Intracranial volume (ICV) segmentation that covers the whole volume inside the skull bone and hippocampal segmentation which is used for example in dementia and in mesial temporal lobe epilepsy (mTLE). (6) Ventricle-brain ratio (VBR) is another brain segmentation application; it is commonly used in psychiatric studies of the brain, it represents the ratio of the ventricle size divided by the whole brain size.
Images obtained with three major MRI pulse sequences are used to segment ICV; proton density PD (7,8), T1 (9), and T2 (10). Another technique uses the combination of PD and T2 (11).
Manual ICV is the reference method to segment total intracranial volume (12). However, automatic ICV software tools are available such as FreeSurfer (http://surfer.nmr.mgh.harvard.edu/) and Statistical Parametric Mapping (SPM) (www.fil.ion.ucl.ac.uk/spm), but it is been reported that those software tools require long time for correlation and approval. (13) (14) The main challenge that faces the segmentation software is whether it produces operator-independent result or not. It is important to assess differences between operators because if the program is robust it can be used as a reference to compare other existing software applications that extract ICV. Smart paint is a semi-automated method (15) that was developed by Filip Malmberg (http://www.cb.uu.se/~filip/) who works at Centre for Image Analysis in Uppsala.
SmartPaint is generic segmentation software that can be used for all types of segmentations in all types of images since it works based on the differences intensity between image voxels. It
6 has been used to segment human prostate from MRI, and it has been using to segment ICV. It uses gradient information to help the user paint inside an object each time a voxel is inside the circle while painting.
The purpose of this study was to compare inter-subject variations of ICV measurements to determine if the volumes are operator independent using the semi-automated method Smart paint, (15).
Material and methods
Subjects
Longitudinal MRI data from the Prospective Investigation of the Vasculature in Uppsala Seniors (PIVUS) study (http://www.medsci.uu.se/pivus/) acquired in 2006-2011 were used for the evaluation of the accuracy and reproducibility of ICV segmentation. Forty Subjects (mean age, 77.5 ±2.5 years) included in this study had no history of neurological disorders or cognitive complaints and had normal neurological examinations. Subject’s data were acquired at age of 75 years or age of 80 years. Informed consent was obtained from all subjects and the study had local ethics committee approval.
Imaging
Scans were performed on the same 1.5 Tesla clinical MRI scanner (Achieva, Philips Healthcare, Best, Netherlands). T2-wieghted sequence has been employed as dual echo sequences, the shorter echo time (TE<30msec) is PD, and the longer echo time (TE>90msec) is T2. Dual echo sequences provide images with different echo time without increasing the scan time. A PD/T2-weighted dual echo sequence was used with 24-cm field of view and 256x256 matrixes. Scans obtained used the following acquisition parameters: Echo time=20.7/100 ms, Repetition time= 3000 ms, flip angle =90 degrees, resolution= 0.94x0.94x3.0 mm. There was an upgrade on the MRI scanner between the scans obtained at age 75 and 80 years, but the scanning protocol was the same for both ages (same echo time, repetition time etc.). PD/T2 dual echo uses two echo times to read out two different images simultaneously PD and T2. PD was used as the boundary between CSF and cortical bone is clearly visible in these images based on the assumption that PD=100% for CSF and PD=50% for bone. Positioning was low in the head coil (toward the feet) to optimize imaging of the cerebral cortex.
7 Image analysis
ICV was measured using SmartPaint, Filip Malmberg (http://www.cb.uu.se/~filip/) (15) (A New Interactive Segmentation Method Applied to Segmentation). SmartPaint is a semi-automated method for ICV segmentation, allows image viewing simultaneously and displays images in axial, coronal, and sagittal orientations. Two operators segmented ICV from 40 PD-weighted MR images blindly to each other. Segmentation is done by multispectral analysis, and evaluation of pixel intensities in structural sequences. The intracranial area was outlined on axial slices to define the border between brain and CSF. A segmentation mask filling 75-80% of ICV was loaded with every scan to save work and time. Then manual correction of these boundaries for every single slice was performed using “seed and expand” to override and adjust the outline when errors occurred such as connections to bone and fat. ICV was segmented by manually tracing the borders of the included regions.
Segmentation included the dural sinuses, the cerebellum even it is below vertebral level C1, and all tissue inside the brain except; the superior sagittal sinus, the internal carotid arteries, the optic nerves, the pituitary gland, and Meckel’s cave (trigeminal cave). ICV inferiorly covers to the cerebellum level when occipital condyles are clearly visible and appear like feet.
Figure 1: Panel of three planes screenshot of SmartPaint segmentation software. the green circle in (A) is the painting software tool to perform seed and expand method, the green line in (B) and (C) represent the Axial slice (A) in the sagittal and coronal respectively.
8 A standard computer (3.40 GHz, 16 GB of RAM) was used by the two observers to perform the segmentation
Statistical analysis
Paired t-test was performed to test for evidence of the manual segmentation differences in mean volumes between the two operators. Person’s correlation coefficient was used to measure the extent of linear association between two operators. Bland Altman plots and Regression analysis were performed to check the agreement between two observers. P values < .05 were considered statistically significant. Intraclass correlation coefficient (ICC) was used to assess the reproducibility of quantitative measurement of the ICV performed by different users (16,17). Coefficient of variance (CV) is given by the following equation; CV= (SD of the ICV differences / mean of the ICV differences) × 100% (18).The statistical analysis was performed with the Statistical Package for the Social Sciences software; Version 12.0 (SPSS, Chicago, Illinois) and Microsoft excel 2010.
Results
Results from intracranial volume segmentation of 40 subjects performed with SmartPaint software by two operators is shown in table 1. The outlined ICV contours are shown in fig1. A paired-sample t-test was used to compare the volume of ICV segmentation (mean) between two operators. There was no significant difference in the mean for observer 1 (m=1468 ml, SD=159) and the observer 2 (m=1459 ml, SD=158) conditions; t (39) =7, p>0 .08.
Table 1: Intracranial volume of SmartPaint method of two operators from 40 subjects.
Operator 1 Operator 2
Mean (ml) 1433 ml 1424 ml
Median (ml) 1433 ml 1426 ml
Range (ml (1205-1870) ml (1205 -1866) ml
The difference between two operators population mean (ml) was ± 9 (ml) indicting very close result
Comparison between two operators showed high correlation shown in figure 2 for both ages. (Age 75: r=.99, P>0.08; age 80: r=0.99, P>0.08). The mean difference between the two operators was -5.99±11.48ml corresponding to typical error of 6.07. Intraclass correlation (ICC) was 0.998 showing high reproducibility of SmartPaint. Coefficient of variation (CV) was 0.907%.
9
Figure 2.Correlation of Rater 1 ICV segmentation corresponding Rater 2 manual method used combined PD and T2. R2 linear interpolation, Pearson’s correlation coefficient squared (the Coefficient of determination).
The differences between operator 1 and operator 2 plotted against their mean for each operator for ICV segmentation subjects, together with the 95% confidence interval (CI) are shown in Figure 3. Note: several subjects have the same value.
y = 0.9941x - 0.9279 R² = 0.9978 y = 0.995x - 1.5654 R² = 0.9955 1190 1290 1390 1490 1590 1690 1790 1890 1190 1390 1590 1790 Rat e r 2 Rater 1
ICV - inter rater variability
Age 75 Age 80
Linear (Age 75) Linear (Age 80)
10
Figure 3: Bland-Altman plot of the sample population. The horizontal dashed lines are the 95% confidence intervals of the difference between the two operators. User 2 indicates operator 2, and user 1 indicate operator 1.
Discussion
We have evaluated the semi-automatic software Smart paint by segmenting the intracranial volume of forty healthy subjects by two operators.
Interclass and Pearson correlations’ were performed, interclass correlation and Pearson correlation showed very close results. and the difference in mean between two segmented ICV was close to zero, and the coefficient of variance was less than 1%.
FreeSurfer and SPM software applications are widely used for ICV; however both tools show overestimation and underestimation respectively (19) (20). FreeSurfer has been reported to give high significant difference compared to the manual method, overestimation might occur due to using T1-weighted MRI sequences in which it is difficult to detect the difference between bone and CSF (19). SPM shows a relative error when compared to the reference manual method (20). Recent study (14) evaluated 19 healthy subjects by comparing new application with SPM5 and manual reference method found underestimation of ICV when using SPM software, the underestimation might refer to the portion of large are of CSF due to radio-frequency field inhomogeneities in MR images.
Difference Plot -35 -30 -25 -20 -15 -10 -5 0 5 10 1200 1300 1400 1500 1600 1700 1800 1900 Mean of All D iffe re n c e (U s e r2 - U s e r1 ) Identity Bias (-9.294) 95% Limits of agreement (-25.786 to 7.198)
11 In our paper we used the MRI sequence PD, the benefits behind this is the ability for user to see the border of brain tissue and the bone edge. The disadvantages of using the T1 alone as evaluated by de Boer (21) is the brain tissue borders is not visible as in the PD, However, the contrast between the WM and the GM is much higher in the T1 images than the PD ones. The PD part of the dual echo sequence added no extra scanning time, the resulting image could be used for any further brain segmentation study or application.
The ICV of adults is unchanged, wheras the brain tissue volume decreases with age. In between the 75 and 80 year scans, there was a major upgrade of the MRI scanner. The scanner upgrade might have influenced the segmentation and thereby the study of the brain tissue difference over time. However, our paper investigated the operator’s segmentation of ICV and assessed the difference and the reliability of the Smart paint. The scanner upgrade did not influence the results we got.
Segmentation time was 15-20 minutes for each subject’s images, and that is labour and time expenditure. There was an option in the software that the user can apply a segmented mask that fills about 75%-80% of the brain tissue, but still work was needed to adjust the segmentation at the tissue borders.
Our study had limitations; it would have been preferable to include more subjects to strengthen the statistical power. A further limitation the ICV segmentation done by only two operators, it would be better if more operators evaluate the segmentation software Smart paint.
Conclusion
Smart paint segmentation results indicate high correlation and accuracy between the two operators; therefore, the software can be used for segmenting ICV.
Acknowledgment I take this opportunity to express my profound gratitude and deep regards to my guide my supervisor Prof. Elna-Marie Larsson and Dr. Richard Nordenskjold for their extremely guidance, monitoring and constant encouragement throughout the ICV project, and for letting me take part in this project.
12
References
1 Buckner RL, Head D, Parker J, Fotenos AF, Marcus D, Morris JC, et al. A unified approach for morphometric and functional data analysis in young, old, and demented adults using automated atlas-based head size normalization: reliability and validation against manual measurement of total intracranial volume. Neuroimage. 2004;23(2).
2 Chan D, Janssen JC, Whitwell JL, Watt HC, Jenkins R, Frost C, et al. Change in rates of cerebral atrophy over time in early-onset Alzheimer's disease: longitudinal MRI study. Lancet. 2003;362(9390):1121-2.
3 Henley SMD, Wild EJ, Hobbs NZ, Frost C, MacManus DG, Barker RA, et al. Whole-Brain Atrophy as a Measure of Progression in Premanifest and Early Huntington's Disease.
Movement Disorders. 2009;24(6):932-6.
4 Chan D, Fox NC, Jenkins R, Scahill RI, Crum WR, Rossor MN. Rates of global and regional cerebral atrophy in AD and frontotemporal dementia. Neurology. 2001;57(10):1756-63.
5 Bartzokis G, Beckson M, Lu PH, Nuechterlein KH, Edwards N, Mintz J. Age-related changes in frontal and temporal lobe volumes in men: a magnetic resonance imaging study. Arch Gen Psychiatry. 2001;58(5):461-5.
6 Akhondi-Asl A, Jafari-Khouzani K, Elisevich K, Soltanian-Zadeh H. Hippocampal volumetry for lateralization of temporal lobe epilepsy: Automated versus manual methods. Neuroimage. 2011;54:S218-S26.
7 Hartley SW, Scher AI, Korf ES, White LR, Launer LJ. Analysis and validation of automated skull stripping tools: a validation study based on 296 MR images from the Honolulu Asia aging study. Neuroimage. 2006;30(4):1179-86.
8 Palm WM, Walchenbach R, Bruinsma B, Admiraal-Behloul F, Middelkoop HA, Launer LJ, et al. Intracranial compartment volumes in normal pressure hydrocephalus: volumetric
assessment versus outcome. AJNR Am J Neuroradiol. 2006;27(1):76-9.
9 Whitwell JL, Crum WR, Watt HC, Fox NC. Normalization of cerebral volumes by use of intracranial volume: Implications for longitudinal quantitative MR imaging. American Journal of Neuroradiology. 2001;22(8):1483-9.
10 Jenkins R, Fox NC, Rossor AM, Harvey RJ, Rossor MN. Intracranial volume and Alzheimer disease: evidence against the cerebral reserve hypothesis. Arch Neurol. 2000;57(2):220-4.
11 Callen DJ, Black SE, Gao F, Caldwell CB, Szalai JP. Beyond the hippocampus: MRI volumetry confirms widespread limbic atrophy in AD. Neurology. 2001;57(9):1669-74. 12 Ambarki K, Wåhlin A, Birgander R, Eklund A, Malm J. MR imaging of brain volumes: evaluation of a fully automatic software. AJNR Am J Neuroradiol. 2011;32(2):408-12.
13 Pengas G, Pereira JM, Williams GB, Nestor PJ. Comparative reliability of total intracranial volume estimation methods and the influence of atrophy in a longitudinal semantic dementia cohort. J Neuroimaging. 2009;19(1):37-46.
13 14 Ambarki K, Lindqvist T, Wåhlin A, Petterson E, Warntjes MJ, Birgander R, et al.
Evaluation of automatic measurement of the intracranial volume based on quantitative MR imaging. AJNR Am J Neuroradiol. 2012;33(10):1951-6.
15 Malmberg F, Lindblad J, Nystrom I. Sub-pixel Segmentation with the Image Foresting Transform. Combinatorial Image Analysis, Proceedings. 2009;5852:201-11.
16 Fleiss JL. Analysis of data from multiclinic trials. Control Clin Trials. 1986;7(4):267-75. 17 Rankin G, Stokes M. Reliability of assessment tools in rehabilitation: an illustration of appropriate statistical analyses. Clin Rehabil. 1998;12(3):187-99.
18 Rudick RA, Fisher E, Lee JC, Simon J, Jacobs L. Use of the brain parenchymal fraction to measure whole brain atrophy in relapsing-remitting MS. Multiple Sclerosis Collaborative Research Group. Neurology. 1999;53(8):1698-704.
19 Lehmann M, Douiri A, Kim LG, Modat M, Chan D, Ourselin S, et al. Atrophy patterns in Alzheimer's disease and semantic dementia: a comparison of FreeSurfer and manual
volumetric measurements. Neuroimage. 2010;49(3):2264-74.
20 Keihaninejad S, Heckemann RA, Fagiolo G, Symms MR, Hajnal JV, Hammers A, et al. A robust method to estimate the intracranial volume across MRI field strengths (1.5T and 3T). Neuroimage. 2010;50(4):1427-37.
21 de Boer R, Vrooman HA, Ikram MA, Vernooij MW, Breteler MM, van der Lugt A, et al. Accuracy and reproducibility study of automatic MRI brain tissue segmentation methods. Neuroimage. 2010;51(3):1047-56.
