Digital neuropathology of neurodegenerative disorders: Foundations, research advances, and future directions

Aaron M. Rosado, Juan C. Vizcarra, Shivam Sharma, C. Dirk Keene, Charles L. White, Ain Kim, Shelley L. Forrest, Gábor G. Kovács, Chen‐Nee Chuah, Margaret E. Flanagan, Thomas M. Pearce, Brittany N. Dugger, David A. GutmanAlzheimer 2025 · DOI

TL;DR

This review explores the burgeoning field of digital neuropathology for neurodegenerative diseases, highlighting how whole slide imaging and AI are revolutionizing the analysis of brain tissue. It details the technologies, challenges, and future directions for leveraging digital pathology to deepen our understanding of diseases like Alzheimer's and improve diagnostic capabilities.

What Problem Does This Paper Address?

Diagnosing neurodegenerative diseases, particularly Alzheimer's Disease (AD) and related dementias (ADRD), relies heavily on post-mortem neuropathological examination, considered the gold standard. However, traditional methods involve laborious manual analysis of physical slides, making large-scale research and data sharing challenging. The sheer volume of data generated by whole slide imaging (WSI) also presents significant storage and computational hurdles. This paper addresses the need to modernize neuropathology workflows by integrating digital technologies, particularly AI and machine learning, to overcome these limitations, enhance analytical capabilities, and accelerate research into the pathogenesis and clinical presentation of these devastating diseases.

Methods

The paper provides a comprehensive overview of digital pathology technologies relevant to neuropathology. This includes:

  • Whole Slide Imaging (WSI): Discusses the evolution of slide scanners, from early slow devices to modern high-throughput systems capable of capturing entire slides in minutes. It touches upon the technical aspects of WSI, including file formats (TIFF, DICOM) and the pyramidal image structure used for efficient viewing.
  • Digital Pathology Software: Reviews both proprietary and open-source software for WSI visualization, annotation, and analysis. This includes tools for basic viewing, complex image analysis, and integration with AI/ML models.
  • Machine Learning (ML) and Artificial Intelligence (AI): Explores the application of ML, particularly deep learning (DL) techniques like Convolutional Neural Networks (CNNs), Generative Adversarial Networks (GANs), and Transformer networks, for tasks such as object detection, segmentation, classification, and quantitative analysis of histopathological features.
  • Computer Vision (CV): Details how CV techniques are used to extract quantitative data from digital images, such as positive pixel counts for staining intensity and cell detection.
  • Data Aggregation and Sharing Platforms: Discusses the infrastructure and tools needed for managing, storing, and sharing large WSI datasets, including cloud-based solutions and open-source platforms like the Digital Slide Archive (DSA).
  • Deidentification Tools: Highlights the importance and methods for removing protected health information (PHI) from WSIs to facilitate data sharing.

Datasets

The paper does not focus on specific novel datasets introduced for training or evaluation. Instead, it emphasizes the need for large, well-curated, and harmonized datasets for neurodegenerative disease research. It references existing data sources like those from AD Research Centers (ADRCs) and the National Alzheimer's Coordinating Center (NACC), and discusses the challenges in aggregating and standardizing data from various brain banks.

Key Insights and Ideas

  • Digital Pathology is Maturing: The field has moved beyond basic digitization to sophisticated AI-driven analysis, offering unprecedented quantitative insights into brain tissue.
  • WSI Data is Massive: A single brain case can generate hundreds of gigabytes of data, posing significant infrastructure and computational challenges that require innovative solutions.
  • AI is Revolutionizing Analysis: ML models can now identify and quantify subtle histopathological features, correlate them with clinical data, and even predict one stain from another (e.g., predicting IHC from H&E-LFB).
  • Collaboration is Crucial: Bridging the gap between computational researchers and domain experts (neuropathologists) is essential for developing clinically relevant and generalizable AI tools.
  • Standardization is Key: The lack of standardized WSI formats and analysis protocols across institutions hinders data sharing and model generalizability.
  • Beyond AD: While AD is a primary focus, the digital pathology approach is applicable to a broader spectrum of neurodegenerative diseases.

Key Results

The paper is a review and does not present novel experimental results with specific quantitative metrics. However, it highlights the potential of digital neuropathology and ML, citing examples of studies that have:

  • Demonstrated the ability of ML models to quantify Aβ plaque burden, correlating with Braak NFT stage and Thal amyloid phase.
  • Developed models for NFT detection and segmentation, as well as for various plaques, microinfarcts, and other histopathological features.
  • Shown that AI can achieve impressive accuracy (AUCs of 0.92) in predicting IHC staining from H&E-LFB images, suggesting that existing data can be leveraged for new insights.
  • Quantified microglial activation and identified distinct microglial and astrocyte phenotypes in AD brains.

Limitations and Future Work

Limitations:

  • Data Volume and Storage: The immense size of WSI files presents ongoing challenges for storage, transfer, and computational processing.
  • Standardization: A lack of standardized WSI formats, staining protocols, and analysis pipelines across institutions hinders data sharing and model generalizability.
  • Generalizability of ML Models: Models trained on data from one institution or scanner may not perform well on data from others due to variations in pre-analytic variables.
  • Cost of Infrastructure: Implementing digital pathology infrastructure, including scanners, storage, and high-performance computing, can be prohibitively expensive for some institutions.
  • Regulatory Hurdles: While FDA approval for WSI is increasing, its adoption for primary diagnosis is still evolving.

Future Work:

  • Development of Foundation Models: Creating large, general-purpose foundation models adapted for neuropathology use cases to improve performance and reduce the need for task-specific training data.
  • Enhanced Data Sharing and Collaboration: Fostering cross-institutional collaborations and developing robust platforms (like the Brain Digital Slide Archive - BDSA) for sharing large WSI datasets.
  • Improved Deidentification Tools: Developing user-friendly, open-source tools for deidentifying WSIs to facilitate data sharing.
  • Middleware Solutions: Creating software infrastructure that bridges WSI scanners, analysis platforms, and AI tools to streamline workflows.
  • Focus on Immune and Inflammatory Contributions: Leveraging digital pathology to better understand the role of immune cells and inflammation in neurodegeneration.
  • Integration with Clinical Data: Developing models that can integrate histopathological data with clinical trajectories, neuroimaging, and fluid biomarkers for more comprehensive patient phenotyping and diagnosis.

Summary updated

Abstract

A neuropathology examination after death remains the gold standard for differentiating between Alzheimer disease (AD) and AD and related dementias (ADRD). Increasing interest and familiarity with digital imaging highlights recent shifts to modernize pathology workflows by leveraging technology that automates imaging and analysis. This review provides an overview of digital pathology technologies and their associated infrastructure, available open-source and proprietary digital pathology software, relevant background on neurodegenerative histopathological features, and computational research. It further examines recent developments in digital pathology in neurodegenerative disease research with an emphasis on machine learning. We discuss evidence supporting how recently developed technologies and methodologies can enhance our understanding of histopathologic features of neurodegeneration and correlations of histopathologic features with cognitive performance and age at death. Finally, we review potential directions for neurodegenerative disease digital pathology research given trends in technological infrastructure development and other digital pathology research. HIGHLIGHTS: Provides a historical summary of digital pathology with respect to neuropathology. Examines key digital pathology technologies. Explores digital pathology applications in neurodegenerative disease and their contribution to research. Discusses the future of digital neuropathology.
BibTeX
@inproceedings{rosado2025digital,
  author = {Aaron M. Rosado and Juan C. Vizcarra and Shivam Sharma and C. Dirk Keene and Charles L. White and Ain Kim and Shelley L. Forrest and Gábor G. Kovács and Chen‐Nee Chuah and Margaret E. Flanagan and Thomas M. Pearce and Brittany N. Dugger and David A. Gutman},
  title = {{Digital neuropathology of neurodegenerative disorders: Foundations, research advances, and future directions}},
  booktitle = {Alzheimer},
  year = {2025},
  doi = {10.1002/alz.70775},
}
Figure