by Christos Evangelou, MSc, PhD – Medical Writer and Editor

In a new study, a multidisciplinary team of researchers at Rush University Medical Center (RADC) developed a digital pathology pipeline to assess Alzheimer’s disease pathologies, such as beta-amyloid plaques and tau neurofibrillary tangles.¹

Digital quantification of beta-amyloid plaques and tau neurofibrillary tangles was efficient, reproducible across magnifications and repeat scans, and recapitulated expected regional patterns of pathology.¹ In addition, the digital pathology data showed high correlations with historical manual and stereology measures, as well as with clinical diagnoses of normal cognition, mild cognitive impairment, and dementia.

This digital pathology pipeline enables high-throughput, reproducible quantification of Alzheimer’s disease pathologies and integration of new data with historical datasets in long-running brain aging studies.

“This study highlights that to successfully implement or transition to a digital pathology workflow requires a multidisciplinary team,” said Alifiya Kapasi, PhD, who is the first author of the study. “Our findings also demonstrate the potential utility of whole slide image analysis to more accurately classify disease burden in postmortem tissue‐based studies.”

The report was published in the Journal of Neuropathology & Experimental Neurology.

Rationale: Overcoming Microscopy Limitations

Historically, the assessment of Alzheimer’s disease pathologies, such as beta-amyloid plaques and tau neurofibrillary tangles, has relied on labor-intensive manual counting and semiquantitative stereology scoring systems.¹ These traditional methods are not only time consuming but also susceptible to interobserver variability and inherent subjectivity. Furthermore, the sampling approaches used in these techniques often fail to capture the full extent of pathology across the entire brain region of interest.¹

“At the Rush Alzheimer’s Disease Center, the Neuropathology Laboratory provides detailed neuropathologic evaluation of postmortem brain tissue for aging and dementia research studies. Due to the challenges associated with manual counting methods, our objective with this study was to use digital algorithms on whole slide images to automate quantification of AD pathologies in a robust and reproducible manner,” said Dr. Kapasi.

Approach: Building a Multidisciplinary, High-throughput Digital Pathology Pipeline

Recognizing the limitations of conventional microscopy, the RADC assembled a multidisciplinary team comprising histologists, neuropathologists, information technology (IT) personnel, statisticians, and database managers.¹ Their mission was to develop a robust digital pathology pipeline that could accurately quantify Alzheimer’s disease pathologies while streamlining workflows and enhancing reproducibility.

By combining the expertise of different disciplines and employing cutting-edge technologies, the team optimized and validated digital algorithms to quantify beta-amyloid load and tau tangle density from whole slide images obtained through high-throughput scanning.¹

“We optimized and validated a positive pixel image analysis algorithm to quantify the percentage of cortex occupied by beta-amyloid positivity and a nuclear image analysis algorithm to quantify tau tangle burden. Importantly, we used statistical approaches to harmonize newly generated digital data with historical microscopy data,” explained Dr. Kapasi. “The unique aspect of this work is the collaboration of experts across multiple disciplines to facilitate transition from traditional microscopy approaches to a digital pathology workflow.”

Validation of Digital Pathology Pipeline

The team optimized and validated their digital pathology pipeline to quantify beta-amyloid load and tau tangle density from whole slide images. Comparison of digital pathology measures and historical manual and stereology data showed a correlation of 0.83 for beta amyloids and 0.94 for tau tangles.¹ These findings underscore the accuracy and reliability of the digital quantification approach.

High Efficiency, Reproducibility, and Clinical Relevance

The digital pathology pipeline demonstrated high efficiency, with quantification taking 5-10 minutes per slide.¹ The digital pathology workflow provided remarkable reproducibility across different magnifications and repeated scans. In addition, the digital measures recapitulated the expected regional patterns of pathology, aligning with established knowledge in the field.¹

The team also found that the digital pathology data exhibited strong associations with clinical diagnoses of normal cognition, mild cognitive impairment, and dementia, as well as neuropathologic diagnoses based on Thal phases (for amyloid) and Braak stages (for tau tangles).¹ These findings underscore the potential of digital pathology to bridge the gap between neuropathological findings and clinical manifestations of Alzheimer’s disease.

Harmonizing Past and Present: Integrating Digital and Historical Data

One of the greatest challenges in implementing digital pathology technologies is ensuring seamless integration with existing data and pathology workflows. To improve the statistical power of histology data on Alzheimer’s disease pathologies, the researchers developed regression models to harmonize the newly generated digital pathology data with historical stereology data spanning multiple large autopsy-based cohort studies.¹ This approach paves the way for seamless integration of future digital findings, ensuring continuity and maximizing the impact of long-term studies.

Commenting on the implications of this approach, Dr. Kapasi said: “This study demonstrates that digital pathology can be utilized as a tool to conduct high-throughput quantification of AD pathology and that new digital data can be statistically merged with historical data in longitudinal clinical-pathologic studies.”

Limitations and Future Directions

Although this study demonstrates the feasibility of using a digital neuropathology workflow and data harmonization approach to facilitate large-scale studies of aging and Alzheimer’s disease, it did not assess the potential impact of changes in antibodies or staining protocols over time, which could affect data harmonization across different cohorts or periods.

Furthermore, the study did not investigate the potential of advanced computational pathology techniques, such as machine learning and deep learning. These AI approaches hold immense promise for identifying nuanced patterns in digital tissue slides, and their potential use to identify tissue morphologies in different Alzheimer’s pathologies merits further investigation.

“Future studies testing the validation framework of using digital algorithms for Alzheimer’s disease pathologic changes on brain tissue from different autopsy cohorts and neuropathology sites will be important,” noted Dr. Kapasi.

Looking ahead, the implementation of digital and machine or deep learning pipelines could aid neuropathologists and researchers in quantifying other pathologic features implicated in Alzheimer’s disease and related dementias, such as inflammation, neuronal loss, vascular pathologies, or white matter changes, as well as classifying nuanced patterns or morphologies of specific proteinopathies.

Moreover, the integration of advanced computational pathology techniques could further enhance our understanding of the complex pathological mechanisms underlying Alzheimer’s disease, potentially paving the way for novel therapeutic targets and diagnostic approaches.

The study was funded by the National Institute on Aging.

References

1. Kapasi A, Poirier J, Hedayat A, et al. High-throughput digital quantification of Alzheimer disease pathology and associated infrastructure in large autopsy studies. J Neuropathol Exp Neurol. 2023;82(12):976-986. doi:10.1093/jnen/nlad086