Imaging-based biomarker for characterizing the structure or function of human or animal brain tissue and related uses and methods

Abstract

The invention relates to novel imaging-based biomarkers for characterizing the structure or function of a human or animal brain. These biomarkers can be a weighted confluency sum score (WCSS) or a percent shielding by brain lesions (SbBL). Methods implementing these biomarkers are also disclosed.

Claims

1. An imaging-based biomarker for characterizing a structure or function of human or animal brain tissue based on an image of the human or animal brain tissue, the image showing at least one brain lesion, wherein the imaging-based biomarker is at least one chosen from the group consisting of a weighted confluency sum score and a percent shielding by brain lesions, wherein the weighted confluency sum score is a sum of weighted confluencies over at least one brain lesion on the image, wherein the weighted confluency sum score is calculated according to formula (I): $\begin{array}{cc}\mathrm{WCSS}=\underset{i}{.Math.}{w}_i.Math.{\mathrm{confluency}}_i& \left(I\right)\end{array}$ wherein the confluency for an i.sup.th brain lesion is calculated according to formula (II) or formula (III): $\begin{array}{cc}{\mathrm{confluency}}_i=\sqrt[3]{\frac{1}{36.Math.}.Math.\frac{{\mathrm{surf}}_i^3}{{\mathrm{vol}}_i^2}}-1& \left(\mathrm{II}\right)\\ {\mathrm{confluency}}_i=\sqrt{\frac{1}{4.Math.}.Math.\frac{{\mathrm{circf}}_i^2}{{\mathrm{area}}_i}}-1& \left(\mathrm{III}\right)\end{array}$ wherein WCSS stands for weighted confluency sum score, i is a summation index running over all or any subset of the brain lesions depicted on the image of the brain tissue, w.sub.i is a weighting factor quantifying the relevance of the i.sup.th brain lesion for a considered application, surf.sub.i represents an estimate of the surface area of the i.sup.th brain lesion, vol.sub.i represents an estimate of the volume of the i.sup.th brain lesion, circf.sub.i represents an estimate of the circumference of the i.sup.th brain lesion, and area.sub.i represents an estimate of the area of the i.sup.th brain lesion, wherein the confluency is a measure of a relation between a surface area of a brain lesion and a volume of the brain lesion or between a circumference of the brain lesion and an area of the brain lesion, and wherein the percent shielding by brain lesions of a brain area is a measure for a fraction of brain lesion areas in a surrounding of a considered brain area.

2. The imaging-based biomarker according to claim 1, wherein the percent shielding by brain lesions of a brain area (A) is computed as the percentage of image voxels or image pixels in a predefined volume or area (B) surrounding the considered brain area (A) that belong to a brain lesion (BL) and calculated according to formula (IV):
SbBL.sub.A=100*V.sub.B(BL)/V.sub.B(IV) wherein SbBL.sub.A stands for the percent shielding by brain lesions of the considered brain area (A), V.sub.B stands for the total number of image voxels or image pixels in a predefined volume or area (B) surrounding the considered brain area (A), and V.sub.B(BL) stands for the number of image voxels or image pixels in B that belong to a brain lesion.

3. The imaging-based biomarker according to claim 1, wherein the image is a magnetic resonance image or a positron emission tomography image or a magnetic particle image.

4. A method for characterizing a structure or function of human or animal brain tissue, comprising the following steps: providing an image of human or animal brain tissue, wherein the image is suited to detect brain lesions on it, detecting at least one brain lesion on the image and delineating its outer contour, for each delineated brain lesion, computing a confluency, the confluency being a measure of a relation between a surface area of a brain lesion and a volume of the brain lesion or between a circumference of the brain lesion and an area of the brain lesion, computing a weighted confluency sum score as a sum of weighted confluencies over all delineated brain lesions, wherein the weighted confluency sum score is calculated according to formula (I): $\begin{array}{cc}\mathrm{WCSS}=\underset{i}{.Math.}{w}_i.Math.{\mathrm{confluency}}_i,& \left(I\right)\end{array}$ wherein the confluency of an i.sup.th lesion is calculated according to formula (II) or formula (III): $\begin{array}{cc}{\mathrm{confluency}}_i=\sqrt[3]{\frac{1}{36.Math.}.Math.\frac{{\mathrm{surf}}_i^3}{{\mathrm{vol}}_i^2}}-1& \left(\mathrm{II}\right)\\ {\mathrm{confluency}}_i=\sqrt{\frac{1}{4.Math.}.Math.\frac{{\mathrm{circf}}_i^2}{{\mathrm{area}}_i}}-1& \left(\mathrm{III}\right)\end{array}$ wherein WCSS stands for weighted confluency sum score, i is a summation index running over all or any subset of brain lesions delineated on the image of the brain, w.sub.i is a weighting factor quantifying the relevance of the i.sup.th brain lesion for a considered application, surf.sub.i represents an estimate of the surface area of the i.sup.th brain lesion, vol.sub.i represents an estimate of the volume of the i.sup.th brain lesion, circf.sub.i represents an estimate of the circumference of the i.sup.th brain lesion, and area.sub.i represents an estimate of the area of the i.sup.th brain lesion, using the weighted confluency sum score to characterize the structure or function of the human or animal brain tissue, the image of which has been analyzed.

5. The method according to claim 4, wherein the weighting factor is different for brain lesions located within different brain regions.

6. The method according to claim 5, wherein if a brain lesion is spread over more than one brain region, the highest weighting factor of the respective brain regions is assigned to this brain lesion.

7. The method according to claim 4, wherein the weighting factor is different for brain lesions that are located within cortical grey matter, periventricular white matter, within deep white/grey matter, within subcortical white matter, or within the brain stem.

8. A method for characterizing a structure or function of human or animal brain tissue, comprising the following steps: providing an image of human or animal brain tissue, wherein the image is suited to detect brain lesions on it, detecting at least one brain lesion on the image and delineating its outer contour, yielding a lesion map, computing a percent shielding by brain lesions for at least one selected brain area, wherein the percent shielding by brain lesions of a selected brain area is a measure for a fraction of a surrounding of the selected brain area belonging to brain lesions, wherein the percent shielding by brain lesions of a brain area (A) is computed as the percentage of image voxels or image pixels in a predefined volume or area (B) surrounding the considered brain area (A) that belong to a brain lesion (BL) and calculated according to formula (IV):
SbBL.sub.A=100*V.sub.B(BL)/V.sub.B(IV) wherein SbBL.sub.A stands for the percent shielding by brain lesions of the considered brain area (A), V.sub.B stands for the total number of image voxels or image pixels in a predefined volume or area (B) surrounding the considered brain area (A), and V.sub.B(BL) stands for the number of image voxels or image pixels in B that belong to a brain lesion, using the percent shielding by brain lesions of the at least one selected brain area to characterize the structure or function of the human or animal brain tissue, the image of which has been analyzed.

9. The method according to claim 8, wherein brain areas for computing their percent shielding by brain lesions are selected according to the following steps providing a second image of the same human or animal brain tissue, wherein the second image is suited to provide different information about brain structure or function than the first image, anatomically co-registering the lesion map with the second image, stereotactically normalizing the second image together with a co-registered lesion map into an anatomical standard space to obtain a normalized second image, comparing the normalized second image with at least one equivalent reference image from at least one reference subject to generate an effect map indicating brain areas in which a property of the second image differs from the reference image, computing the percent shielding by brain lesions for each brain area on the effect map.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) Aspects of the invention will be explained more detail with respect to Figures and exemplary embodiments.

(2) FIG. 1A shows a transversal slice of a FLAIR-MR image without delineation of subcortical hyperintensities.

(3) FIG. 1B shows a transversal slice of a FLAIR-MR image with delineation of subcortical hyperintensities.

(4) FIG. 2A shows transversal slices of an anatomical map of a brain.

(5) FIG. 2B shows slices from a T1-weighted MR image corresponding to the slices of FIG. 2A for anatomical orientation.

(6) FIG. 3 shows the results of a computer simulation of spheres composed of a varying number of cubic voxels.

(7) FIG. 4A shows the results of a computer simulation of a confluency sum score of differently shaped lesions in transversal view.

(8) FIG. 4B shows the results of a computer simulation of a confluency sum score of differently shaped lesions in coronal view.

(9) FIG. 5 shows the confluency score of cuboids having different lengths.

(10) FIG. 6 shows a receiver-operating characteristic (ROC) curve of the WCSS for differentiation between patients with vascular cognitive decline and patients without relevant cerebrovascular disease.

(11) FIG. 7A shows FDG PET images of the brain in a patient with Alzheimer's disease.

(12) FIG. 7B shows FDG PET images of the brain in a healthy subject.

(13) FIG. 8A shows FDG PET images of the brain in a patient who was suspected to have Alzheimer's disease.

(14) FIG. 8B shows T1-MR images of the same patient as in FIG. 8A.

(15) FIG. 9A shows a parametric hypometabolism map overlaid to FDG PET images of the brain of a patient.

(16) FIG. 9B shows a parametric SbWMH map and a WMH lesion map overlaid to FDG PET images of the brain of the same patient as in FIG. 9A.

(17) FIG. 10 shows schematically a brain area (denoted as A) shielded by a brain lesion (denoted as BL) in an area (denoted as B) surrounding the considered brain area (A).

DETAILED DESCRIPTION

(18) FIG. 1A shows a transversal slice of a FLAIR-MR image without delineation of subcortical hyperintensities. Such transversal slices are known from prior art. FIG. 1B shows a transversal slice of a FLAIR-MR image with delineation of subcortical hyperintensities. In this Figure, a large confluent lesion can be well distinguished from a small spherical lesion.

(19) A delineation as shown in FIG. 1B is used in a method according to the first exemplary embodiment explained in the following. FIGS. 2A to 6 will be explained with respect to this first exemplary embodiment. FIGS. 7A to 10 will be explained with respect to a second exemplary embodiment.

First Exemplary Embodiment: Calculation of a Weighted Confluency Sum Score (WCSS)

(20) The exemplary embodiment relates to a (computer) system for fully automatic determination of a weighted confluency sum score (WCSS). This system utilizes magnetic resonance (MR) image data of the human brain. It starts with the automatic detection of all brain lesions in the MR image and accurate delineation of their outer contours. An exemplary result is shown in FIG. 1B. The system implements an algorithm for automatic detection of FLAIR-hyperintense white-matter lesions which has been proposed by Schmidt and co-workers for application in with multiple sclerosis [3].

(21) This Schmidt algorithm generates a three-dimensional hyperintensity map which is then binarized. The binarized hyperintensity map is then clustered into separate hyperintensity lesions using the routine spm_bwlabel from the Statistical Parametric Software package (version SPM8, http://www.fil.ion.ucl.ac.uk/spm/). This routine labels connected components on the basis of a connectivity criterion to be specified. Six adjacent voxels (on the surface) have been defined here as connectivity criterion.

(22) Then the system computes the confluency for each brain lesion according to formula (II)

(23) 0$\begin{array}{cc}{\mathrm{confluency}}_i=\sqrt[3]{\frac{1}{36.Math.}.Math.\frac{{\mathrm{surf}}_i^3}{{\mathrm{vol}}_i^2}}-1,& \left(\mathrm{II}\right)\end{array}$

(24) Surface and volume of the hyperintensity lesion are computed by counting of voxels as defined in the clustered hyperintensity map. This is computationally very efficient.

(25) The weighting factor w.sub.i for a given hyperintensity lesion is defined according to its localization within the brain: w.sub.i=1, 2, 3, 4 if the lesion is located within periventricular white matter, deep white/grey matter, subcortical white matter, or within the brain stem, respectively. The assignment of the lesion to these four different regions is based on an anatomical map that has been previously created from tissue probability maps provided by SPM8. This anatomical map is depicted in FIG. 2A, wherein subcortical white matter is depicted in dark red, deep white/grey matter is depicted in green, periventricular white matter is depicted in orange and brainstem is depicted in blue. For a better anatomical orientation, FIG. 2B shows corresponding slices from a T1-weighted MR image. If a hyperintensity lesion is located in more than one of the regions, it is assigned the highest weighting factor of these lesions.

(26) Finally, the weighted confluency sum score (WCSS) is computed according to formula (V)

(27) $\begin{array}{cc}\mathrm{WCSS}=\underset{i}{\overset{m}{.Math.}}{w}_i.Math.{\mathrm{confluency}}_i& \left(V\right)\end{array}$

(28) The individual parameters have the same meaning as in case of formula (I). The only difference between formula (I) and (V) is that in case of formula (V) m is used as number of the analyzed brain lesions. Thereby, m refers to the total number of hyperintensity lesions in the hyperintensity map consisting of at least 100 voxels.

(29) The instantly described system has been successfully validated by the following experiments: The algorithm proposed by Schmidt and co-workers for FLAIR-hyperintensity lesions in multiple sclerosis was successfully validated in 44 elderly patients (mean age 80 years) with unclear cognitive impairment from several wards for geriatric inpatients. As already explained above, there is no perfect sphere in MR images, but only edgy approximations of a sphere composed of cubic voxels. Computer simulations of spheres composed of a varying number of cubic voxels showed that the resulting error in the confluency can be neglected for spheres composed of at least 100 voxels. The according results are shown in FIG. 3. For spheres composed of at least 100 voxels, the confluency approaches zero, i.e. the value of an ideal sphere. Computer simulations were performed to show that the confluency according to formula (II) indeed is a useful measure of confluency of brain lesions. Specifically, 6 spherical lesions of 10 mm radius each and one cuboid simulating the confluency of the 6 spheres to one single contiguous lesion were analyzed. The results are depicted in FIGS. 4A and 4B. The calculated WCSS was almost zero for the pattern consisting of the 6 spherical lesions, whereas it was only 0.74 for the cuboid (all weighting factors were set to 1). When the number of spherical lesions that confluenced to a cuboid was increased, the WCSS of the cuboid showed a continuous increase. The according results can be seen in FIG. 5 showing that the confluency score of a cuboid increases continuously with its length, i.e. the number of spherical lesions that confluenced to the cuboid. The WCSS of the pattern of spherical lesions remained almost zero, independent on the number of spherical lesions (all weighting factors were set to 1). In a clinical evaluation, the area under a receiver-operating characteristic curve for differentiation between patients with vascular cognitive decline and patients without relevant cerebrovascular disease by the WCSS was 0.830. This is shown in FIG. 6. This clearly shows that the WCSS is clinically useful.

Second Exemplary Embodiment: Calculation of a Percent Shielding by White Matter Hyperintensities (SbWMH)

(30) Patho-physiological changes in the brain caused by neurodegenerative diseases such as Alzheimer's disease include alterations of brain activity (synaptic dysfunction). Positron emission tomography of the brain with the glucose analog F-18-fluorodeoxyglucose (FDG PET) provides biomarkers for (synaptic) function and dysfunction, as depicted in FIGS. 7A and 7B.

(31) FIG. 7A shows FDG PET images of the brain in a patient with Alzheimer's disease. These images show a reduction of brain activity compared to a healthy subject (cf. FIG. 7B) in most brain regions except visual and motor cortex, subcortical brain structures and the cerebellum. The reduction is most pronounced in posterior cingulum/precuneus area and the parietotemporal cortex (indicated by arrows). This pattern is typical for Alzheimer's disease. The reduction of brain activity is mainly caused by reduced synaptic activity. Although there is some loss of brain tissue (atrophy) in Alzheimer's disease, its impact on brain FDG PET is rather small, at least at early stages of the disease. In case of strong atrophy, the effect on FDG PET can be taken into account by partial volume correction.

(32) In old patients, however, the detection of synaptic dysfunction associated with neurodegenerative disease is complicated by the high rate of vascular co-morbidity, for example infarcts of the brain of varying size. This is depicted in FIGS. 8A and 8B. FIG. 8A shows brain FDG PET images in a patient who was suspected to have Alzheimer's disease. The pattern of reduction in the PET indeed looks rather similar to the typical pattern in Alzheimer's disease (cf. FIG. 7A). However, inspection of the MRI of the same patient (depicted in FIG. 8B) reveals several infarcts and strong white matter disease (indicated by arrows). This vascular pathology fully explains the abnormal findings in the FDG PET. Therefore, there is no indication of Alzheimer's disease in this patient. The patient has vascular cognitive decline.

(33) It is evident that there is no FDG uptake in infarcted tissue (scar). Whether or not a reduction of FDG uptake is the direct consequence of an infarct can be tested rather easily by coregistering T1- and/or T2-weigthed MRI (in which most infarcts are clearly displayed) to the FDG PET. However, not only infarcts but also impairment of axonal connections can cause reduced synaptic activity in both neighboring and distant grey matter regions, due to interruption of axonal tracts to this region.

(34) Since white matter hyperintensities are to be considered as specific brain lesions, the novel biomarker SbWMH is an embodiment of the biomarker SbBL (shielding by brain lesions). It is a marker of impairment of axonal connections in form of a percent shielding of cortical brain regions by white matter hyperintensities.

(35) A processing pipeline for fully automated computation and display of SbWMH has been implemented as a MATLAB script. For some processing steps, tools from the statistical parametric mapping software package are used (version SPM8). The pipeline comprises the following steps.

(36) Extraction of White Matter Hyperintensities from Structural MRI

(37) The Lesion Segmentation Toolbox, a freely-available add-on to SPM8, is used to extract WMHs from the patient's structural MRI. The toolbox requires a high-resolution T1-weighted MRI and a FLAIR-MRI as input. The output is a binary lesion map delineating WMHs in the patient's native space.

(38) Co-Registration and Spatial Normalization of Lesion Map and FDG PET

(39) SPM's co-register tool is used to register the lesion map with the FDG PET. SPM's normalize tool is used to transform co-registered images into the anatomical space of the Montreal Neurological Institute (MNI).

(40) Generation of Hypometabolism Map

(41) A (homoscedastic) t-test for two independent samples is used to compare the patient's normalized FDG PET to the normalized FDG PETs of a database of aged-matched healthy controls. The global FDG-uptake is used as reference value for intensity scaling prior to the statistical test. Reduced scaled FDG-uptake is defined as hypometabolism if p0.001. This results in a parametric map of hypometabolism. Such a hypometabolism map is shown in FIG. 9A depicting a parametric hypometabolism map (blue blobs) overlaid to the patient's FDG-PET.

(42) Voxel-Wise Computation of SbWMH

(43) The SbWMH is computed for each hypometabolic voxel as the fraction of neighboring white matter voxels affected by WMH (FIG. 10). The 50 ml white matter voxels closest to the hypometabolic voxel are used as white matter neighborhood. White matter is defined by a binary mask that has been generated from the a priori tissue probability maps used for WMH lesion segmentation. The SbWMH values are saved to a 3-dimensional parametric map.

(44) Display

(45) The slover tool as implemented in SPM8 is used to display the SbWMH map (as blobs with jet colortable) together with the WMH lesion mask (as contours) superimposed to the patient's FDG PET in MNI space. An according map is shown in FIG. 9B depicting a parametric SbWMH map (jet-colored blobs) and WMH lesion map (red contours) overlaid to the FDG PET. The SbWMH values are quantitative: SbWMH=50 means that as much as 50% of the closest 50 ml white matter voxels are affected by WMH. As a consequence, the hypometabolism in this voxel most likely is caused by neighboring WMH, i.e. the hypometabolism is due to cerebrovascular disease, whereas no indication of neurodegenerative disease could be found.

(46) In the example depicted in FIGS. 9A and 9B, the hypometabolism in the left lateral frontal cortex, the left parietotemporal cortex and in the precuneus can be explained by WMH (green arrows). The hypometabolism in the medial frontal cortex most likely is not caused by WMH, it rather might be an unspecific effect of old age (red arrow). The patient most likely does not suffer from Alzheimer's disease (AD), although the pattern of hypometabolism is similar to the typical AD pattern. Thus, in this case, the SbWMH map reduces the risk of misinterpretation the structural alterations of the brain as AD.

(47) The basic idea underlying the percent shielding by brain lesions is illustrated in FIG. 10. The percent shielding by brain lesions of a brain area A is computed as the percentage of image voxels or image pixels belonging to a brain lesion BL in a predefined volume or area B surrounding the considered brain area A. There can be more than one brain lesion BL in the volume or area B all of which contribute to the percent shielding of A.

LIST OF REFERENCES CITED IN THE PRECEDING SECTIONS

(48) 1. Wardlaw, J. M., et al., Neuroimaging standards for research into small vessel disease and its contribution to ageing and neurodegeneration. Lancet Neurol, 2013. 12(8): p. 822-38. 2. Wahlund, L. O., et al., A new rating scale forage-related white matter changes applicable to MRI and CT. Stroke, 2001. 32(6): p. 1318-22. 3. Schmidt, P., et al., An automated tool for detection of FLAIR-hyperintense white-matter lesions in Multiple Sclerosis. Neuroimage, 2012. 59(4): p. 3774-83. 4. Kapeller, P., et al., Visual rating of age-related white matter changes on magnetic resonance imaging: scale comparison, interrater agreement, and correlations with quantitative measurements. Stroke, 2003. 34(2): p. 441-5. 5. Prins, N. D., et al., Measuring progression of cerebral white matter lesions on MRI: visual rating and volumetrics. Neurology, 2004. 62(9): p. 1533-9. 6. van den Heuvel, D. M., et al., Measuring longitudinal white matter changes: comparison of a visual rating scale with a volumetric measurement. AJNR Am J Neuroradiol, 2006. 27(4): p. 875-8. 7. Hernandez Mdel, C., et al., New multispectral MRI data fusion technique for white matter lesion segmentation: method and comparison with thresholding in FLAIR images. Eur Radiol, 2010. 20(7): p. 1684-91. 8. Ramirez, J., et al., Lesion Explorer: a comprehensive segmentation and parcellation package to obtain regional volumetrics for subcortical hyperintensities and intracranial tissue. Neuroimage, 2011. 54(2): p. 963-73. 9. Apostolova, I., et al., Quantitative assessment of the asphericity of pretherapeutic FDG uptake as an independent predictor of outcome in NSCLC. BMC Cancer, 2014. 14: p. 896. 10. Apostolova, I., et al., Asphericity of pretherapeutic tumour FDG uptake provides independent prognostic value in head-and-neck cancer. Eur Radiol, 2014. 24(9): p. 2077-87. 11. Hofheinz, F., et al., Increased evidence for the prognostic value of primary tumor asphericity in pretherapeutic FDG PET for risk stratification in patients with head and neck cancer. Eur J Nucl Med Mol Imaging, 2014. 12. Kochunov, P., Ramage, A. E., Lancaster, J. L., Robin, D. A., Narayana, S., Coyle, T., Royall, D. R., Fox, P., 2009. Loss of cerebral white matter structural integrity tracks the grey matter metabolic decline in normal aging. Neuroimage 45, 17-28. 13. Tullberg, M., Fletcher, E., DeCarli, C., Mungas, D., Reed, B. R., Harvey, D. J., Weiner, M. W., Chui, H. C., Jagust, W. J., 2004. White matter lesions impair frontal lobe function regardless of their location. Neurology 63, 246-253. 14. Reed, B. R., Eberling, J. L., Mungas, D., Weiner, M., Kramer, J. H., Jagust, W. J., 2004. Effects of white matter lesions and lacunes on cortical function. Arch. Neurol. 61, 1545-1550. 15. Glodzik L, Kuceyeski A, Rusinek H, Tsui W, Mosconi L, Li Y, Osorio R S, Williams S, Randall C, Spector N, McHugh P, Murray J, Pirraglia E, Vallabhajosula S, Raj A, de Leon M J. Reduced glucose uptake and A in brain regions with hyperintensities in connected white matter. Neuroimage. 2014 Oct. 15; 100:684-91