Stereology and automated measurement of the human brain
MetadataShow full item record
Stereology supplies image sampling rules to estimate geometric quantities such as volume, surface area, feature length and number. The method is well suited to non-invasive image acquisition methods such as Magnetic Resonance Imaging (MRI). Meanwhile, in Magnetic Resonance (MR) images analysis area, automated software packages have been continuously developed and become well-established tools especially in human brainMR images processing. The aims of the thesis are (1) to combine proper rules to sample MR images with automated or semi-automated data acquisition methods, in order to implement four different design unbiased stereological volume estimators in the study of the human brain, and (2) to compare volume estimates with those obtained from automated software packages.In volume estimation of three-dimensional (3D) objects, besides one traditional stereological method (i.e. the CAVALIERI method), in recent years a number of newly design-based unbiased methods have been published, which include three used in this thesis (i.e. the ISOTROPIC CAVALIERI (ICAV), INVARIATOR (INV) and DISCRETIZED NUCLEATOR (DN) methods). The ICAV and INV methods both allow the estimation of surface area too. The ICAV method enables volume estimation to be unbiased and precise in individual objects while the INV and DN methods make it efficient to estimate the mean volume of a big cohort. To make it be practical in estimating the volume of human brain solely from MR images, in the thesis the ICAV, INV and DN methods were given two operating protocols for rotation and measurement on a commercial software (i.e. ANALYZE) and were performed in a fetal brain study. The ICAV, INV and DN methods were also programmed in three scripts for rotation, gridding and measurement purposes respectively using three freely available software packages (i.e. FSL, R and IMAGEJ), which were applied in three adult brain studies.In volume estimation of three-dimensional (3D) objects, besides one traditional stereological method (i.e. the CAVALIERI method), in recent years a number of newly design-based unbiased methods have been published, which include three used in this thesis (i.e. the ISOTROPIC CAVALIERI (ICAV), INVARIATOR (INV) and DISCRETIZED NUCLEATOR (DN) methods). The ICAV and INV methods both allow the estimation of surface area too. The ICAV method enables volume estimation to be unbiased and precise in individual objects while the INV and DN methods make it efficient to estimate the mean volume of a big cohort. To make it be practical in estimating the volume of human brain solely from MR images, in the thesis the ICAV, INV and DN methods were given two operating protocols for rotation and measurement on a commercial software (i.e. ANALYZE) and were performed in a fetal brain study. The ICAV, INV and DN methods were also programmed in three scripts for rotation, gridding and measurement purposes respectively using three freely available software packages (i.e. FSL, R and IMAGEJ), which were applied in three adult brain studies.A fetal brain study was carried out to test the application of the ICAV, INV and DN methods. Ten fetuses from three maternal backgrounds (i.e. five healthy, three maternal psychological stress and two maternal substance misuse) were scanned in MRI at both the second and third trimesters of pregnancy. Then fetal brain images were motion corrected using SLIMMER software. Volumes of brain parenchyma (the functional tissue of the brain which is made of two types of brain cells, namely neurons and glia) including ventricles were estimated by the ICAV method in isotropic (i.e. having no preferred orientation) and uniformly random (i.e. uniformly distance (interval) apart) (IUR) triplet of orthogonal section planes (i.e. the ICAV ortrip method) and by the INV and DN methods in isotropically random (IR) triplet of orthogonal section planes through a fixed pivotal point (i.e. the INV ortrip and DN ortrip methods). Due to observation of artefacts in MR images and manual input in the methods, inter- and intra-rater reliability studies were performed to investigate both point counting for the ICAV method and segment length measurement for the INV and DN methods among three raters on five fetal brains from the second and five from the third trimesters. Surface area was also estimated using the ICAV method for error prediction. High reliability (Pearson’s r > 0:997) was shown in inter- and intra-rater studies. In both the second and third trimesters, there were no significant difference in mean volumes of all ten brains estimated by the three methods (p > 0:1). For individual estimates, The predicted coefficients of error (CEs) for the ICAV method were 1:5% ± 0:1% in the second trimester and 2:1% ± 0:1% in the third trimester. Basing on one IR section plane for each data, empirical CEs for the INV method in both trimesters were 19:4% ± 2:9% and 18:5% ± 11:6%, and were 21:4% ± 4:5% and 24:1% ± 11:7% for the DN method. CEs could be decreased to 8.1%, 5.5% for the INV ortrip method and 10.3%, 9.7% for the DN ortrip method in both trimesters. This study showed the ICAV method performed precisely in individual volume measurements while the INV and DN methods worked efficiently in population mean volume estimation. Clinically, no significant differences (p > 0:05) of fetal brain volumes among three maternal groups were detected due to small sample size although potentially in comparison with normal fetal brain, volume might be bigger in the maternal stress group and might be smaller in the substance abused group.As the CAVALIERI method is a design-unbiased method, the main source of potential bias (i.e. if a biased method is applied the mean of the estimated values deviates significantly from the true value) will come from observers in operation who would be the author in this PhD study. To examine whether there is bias caused by the author’s manual point counting procedure in the CAVALIERI method, a slice-by-slice comparison on one adult brain volume estimation on MR images between the CAVALIERI method and an automatically reconstructing software (i.e. FREESURFER) was performed. One healthy elderly (male, age 71) brain MRI scan with good image quality was selected from a dataset of 40 patients affected by the ALZHEIMER’s disease (AD) and 22 healthy elderly volunteers. FREESURFER was used to perform individual volumetric analysis on the adult brain automatically, which outlined grey matter and white matter in the cerebrum on each MR image slice. The CAVALIERI method was applied to a series of coronal images obtained with random starting position and at 1 cm intervals from the TALAIRACH transformed and intensity normalized 3D MR image (i.e. nu.mgz) displayed with the compartment boundaries identified by the FREESURFER pipeline suppressed. The uniformly random (UR) test system for point counting was superimposed on each image. The CAVALIERI method in combination with point counting strategy was used to estimate the volume of cerebrum excluding ventricles (the sum of the two cerebral hemispheres including blood vessels and meninges) using EASYMEASURE software on the grey scale MR images. Additionally, the author overlay the brain boundary segmented by FREESURFER on these selected test points and reassessed the images to compute two scores, namely (i) the total number of test points which had been counted but which were seen to lie outside the FREESURFER segmentation and (ii) total number of the new test points that now needed to be additionally included as lying within the FREESURFER segmentation.. The cerebral volume was 972 cm3 estimated by FREESURFER and 960 cm3 by the author using the CAVALIERI method with CE of 0.34%. FREESURFER had 1.3% bigger measure than that estimated by the author. FREESURFER aided point counting estimate was between 948 cm3 and 982 cm3 with mean volume of 965 cm3. The ratio of points counted by the author but were outside the pial boundary segmented by FREESURFER to total points number counted was between 2.6% to 4.0%, and the ratio of test points not counted by the author but were inside the pial boundary segmented by FREESURFER to total points number counted was between 2.7% to 4.9%. Therefore the author’s estimate was in the range of FREESURFER aided estimation by the CAVALIERI method and both volume ratios were close to each other. No bias could be found between the author using the CAVALIERI method and FREESURFER, which gives the author confidence in performing following studies.Furthermore, volume difference of cerebrum excluding ventricles between AD patients and healthy people were investigated using four stereological methods (i.e. the CAVALIERI, ICAV, INV and DN methods) and the FREESURFER software. From the same dataset of 40 AD patients and 22 healthy elderly volunteers, brain MR images of 13 patients and 13 volunteers were selected with good image quality. Inter-reliability and intra-repeatability studies were performed by two observers on three AD and three normal ageing brains. Both the inter-reliability and intra-repeatability studies showed good consistency. There was no significant difference of individual measures among the CAVALIERI and ICAV methods and FREESURFER. The average time taken for each cerebral volume estimation was less than 15 mins by each of the CAVALIERI, ICAV, INV and DN methods. Clinically, the cerebral volume was significantly smaller in the AD patients, which were found using both the CAVALIERI (p = 0:01) and ICAV (p < 0:01) methods and FREESURFER (p = 0:01) although the INV and DN methods were not able to detect this difference. In this adult brain group study, the volume estimates from the CAVALIERI and ICAV methods were competitive with those obtained from FREESURFER, while the INV and DN methods might be more useful if being applied with a larger sample size or the INV ortrip and DN ortrip methods were applied.Lastly, a systematic investigation on potential imaging biomarkers for AD was performed by the FREESURFER software, one of the imaging biomarkers (i.e. volume ratio of cerebrum excluding ventricles to intra-cranium (the contents of the skull above the level of the foramen magnum)) was re-examined by the INV method manually. From the same dataset of 40 AD patients and 22 healthy elderly volunteers, the brain MR images of 27 AD patients and 16 healthy elderly controls between the maximal common age scope of 47 to 71 were selected, which were analysed by FREESURFER for each brain region. Furthermore, to see the effect of AD on normal ageing atrophy, the difference of volume ratios of each brain region to whole brain between AD patients and healthy controls was investigated. Volume ratios of many brain parenchymal regions (e.g. hippocampus (left p = 0:002, right p < 0:001), amygdala (left p < 0:001, right p < 0:001), accumbens area (left p = 0:002, right p = 0:001), left putamen (p = 0:037) and corpus callosum (mid anterior p = 0:03, mid posterior p = 0:032)) to whole brain were found smaller in AD patients while volume ratios of ventricles (both sides of lateral, inferior lateral and 3rd ventricles, p < 0:001) to whole brain were bigger in AD patients. Besides, the INV method was able to detect significant difference of volume ratio of cerebral parenchyma to intra-cranium between AD patients and healthy elderly subjects too (p < 0:01). In comparison with normal ageing-related atrophy in healthy subjects, brain atrophy with ageing in AD patients presented in a different pattern in volume ratios (e.g. right (p = 0:029) and total cortex (p = 0:013) to brain, total grey matter to brain (p = 0:01), cerebral white matter to brain (p = 0:003), cerebellar cortex to brain (left p = 0:019, right p = 0:032), 5th ventricle to brain (p = 0:048), left fimbria to left hippocampus (p = 0:015), left hippocampal-amygdaloid transition area (HATA) to left hippocampus and right presubiculum to right hippocampus (p = 0:016)). In this AD study, many volume ratios of brain regions to whole brain or other brain regions were found different in AD patients and especially volume ratio of cerebral parenchyma to intra-cranium showed potentiality as an imaging biomarker for AD. Ageing atrophy pattern was found different in AD patients too.In conclusion, by programming in freely available R, FSL and IMAGEJ software packages, the CAVALIERI, ICAV, INV and DN methods were able to be performed conveniently and efficiently on human brain volume estimation using MRI. We made the first applications on the volume estimation of fetal brains, healthy brains and brains affected by AD using the ICAV, INV and DN methods. The volume estimates were competitive with those obtained from automated programme (i.e. FREESURFER).