Machine Learning in Digital Neuropathology: Towards a Large-Scale Analysis Platform for Federated Cohorts
Author(s)
Vizcarra, Juan Carlos
Advisor(s)
Gutman, David A.
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Millions of people suffer from neurodegenerative diseases, including Alzheimer’s disease, Parkinson’s disease, and related disorders. Post-mortem analysis of brain tissue is essential in improving our understanding of the underlying biological mechanisms of neurodegeneration. Modern techniques allow digitization of brain tissue glass slides into large images that are rich in data for computational analysis. Developing effective image analysis tools for these datasets is challenging because datasets vary widely, are rarely reported completely, and need to be developed with the expert user, neuropathologist, in mind.
To address these concerns, this work aimed at investigating the intersection between neuropathology, modern computational analysis and data management. In Aim 1, an inter-rater and inter-annotator study was conducted to measure the variability amongst experts and novices in two important tasks related to Alzheimer’s disease pathology. Aim 2 explored the ability to utilize modern computational approaches in machine learning to perform the tasks in Aim 1 and show that even with imperfect ground truth, computational approaches can mimic and perform similar to experts in the field. Finally in Aim 3, a suite of tools, including the NeuroTK platform, were developed to provide all the necessary tools for neuropathologists to run novel large scale image analysis studies. The results of this work highlight the impact of machine learning in neuropathology and provide a suite of powerful open-source tools that will open up large scale computational analysis of digital imaging datasets.
Sponsor
Date
2024-02-01
Extent
Resource Type
Text
Resource Subtype
Dissertation