This Science Short, written by Pulkit Khandelwal, summarizes a manuscript currently under review at Imaging Neuroscience, with a preprint available at https://arxiv.org/abs/2303.12237.
Dementia refers to the loss of memory, language, problem-solving and other thinking abilities that interferes with daily life activities. Dementia can be caused by several neurodegenerative diseases such as Alzheimer’s disease, Lewy body dementia, and frontotemporal dementia, amongst many others. These neurodegenerative diseases are characterized by multiple pathological processes jointly contributing to neurodegeneration in most patients. For example, Alzheimer’s disease (AD) pathology is characterized by two main proteins called tau tangles and beta-amyloid plaques, which are thought to lead to neurodegeneration and cognitive decline. But many patients diagnosed at autopsy with AD also exhibit other vascular diseases or proteins that cannot be reliably detected by imaging live patients. This makes it difficult for clinicians to determine to what extent cognitive decline in individual patients is driven by AD vs. other factors. Clinicians rely upon biomarkers, which are quantifiable characteristics that can help detect and track diseases. For example, it is known that the thickness of the cortical gray matter (the outer spaghetti-like brain structure) decreases over time in patients diagnosed with AD. So, a clinician can look at the thickness of the cortex and judge the prognosis of AD in a patient. The recent modest successes of AD treatments in clinical trials make it even more important to derive biomarkers that can detect and quantify mixed pathology, so that treatments can be prioritized for those most likely to benefit from them.
Associations between pathologies and measures of neurodegeneration, such as thickness and volume of the cortical gray matter or the subcortical areas, can identify patterns of neurodegeneration linked to specific pathologies. Such patterns can then help us create specific biomarkers for early detection of the disease. For example, as mentioned earlier, it is known that a decrease in cortical thickness is associated with Alzheimer’s disease, and accumulation of proteins in the medial temporal lobe region is linked with early onset of AD. These associations were found using imaging in live (antemortem) patients, but our understanding of biomarkers can be greatly aided by examining the brain tissue after death, aka, postmortem. After the autopsy of a person who has succumbed to AD, we can look at the brain tissue under the microscope (histology) and also obtain high-resolution magnetic resonance imaging (MRI) scans. Both techniques can help us understand the build-up of different pathological proteins contributing to the disease. Postmortem MRI is advantageous because it allows imaging at much greater resolution than antemortem and thereby allows structure-pathology associations to be examined with greater granularity and helps us visualize detailed and intricate neuroanatomy.
Given the rising use of high resolution postmortem MRI in neurodegenerative disease research, automated techniques are imperative to effectively analyze such growing datasets. However, many of the approaches focus on antemortem MRI, and there is limited work on developing automated methods for postmortem MRI analysis. Therefore, in this landmark study, we developed a mathematical framework and a set of software tools to produce accurate delineations of the cortical gray matter; subcortical structures (caudate, putamen, globus pallidus, thalamus), white matter (WM) and white matter hyperintensities (WMH) in a set of 135 high resolution 7 tesla postmortem MRI scans of whole brain hemispheres. In other words, we developed a way to look at specific brain regions in high-resolution MRI. This data is part of several ongoing and previous clinical research programs at various laboratories within Penn Medicine and Penn Engineering where dementia patients are treated and then followed to autopsy after death to obtain imaging and pathology. This heterogeneous cohort spanned patients with a variety of diseases such as AD, amyotrophic lateral sclerosis, cerebrovascular disease, Lewy body disease, frontotemporal lobar dementia to name a few. See the paper for more details.
We studied regional patterns of associations at several anatomical locations between cortical thickness and the underlying neuropathology ratings of proteins such as tau and amyloid-beta, and measures of neuronal loss, including Braak staging (a measure to track the progress of AD). As expected, we found that decrease in cortical thickness and volume of the studied brain regions were associated with an increase in the protein build-up, an increase in neuronal loss, and higher Braak stages in areas associated with AD, such as the subregions of the hippocampus and medial temporal lobe. Therefore, postmortem MRI will be helpful for validating and refining measures derived from antemortem studies. Using the developed methods and tools, postmortem MRI will allow future work to study links between structural changes and other local processes beyond pathology, including inflammatory markers and gene expression.
Pulkit Khandelwal is a 5th year Bioengineering PhD student in the Penn Image Computing and Science Laboratory at the University of Pennsylvania under the supervision of Prof. Paul Yushkevich. He is interested in medical image analysis particularly in image segmentation and registration problems to aid image-guided surgery and develop computational methods and tools to understand neurodegenerative diseases. Previously, he completed his masters in computer science in the Shape Analysis Group at McGill University under the supervision of Prof. Kaleem Siddiqi and Prof. Louis Collins.