New Digital Pathology Pipeline Enables Multiregional Quantification of Tau-Induced Damage in Postmortem Brains

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

In a new study, researchers at the University of Cambridge and Bispebjerg University Hospital developed a digital quantification pipeline using machine learning to quantify different types of tau pathology in postmortem brain samples from patients with progressive supranuclear palsy (PSP), a neurodegenerative tauopathy.1

The machine learning pipeline demonstrated high accuracy in quantifying tau pathology in digital slides of postmortem brains from patients with PSP.1 By automating and standardizing quantification, the pipeline may facilitate the investigation of tau distribution patterns and clinicopathological correlations in tauopathies at a much larger scale than feasible manually.

“Our study highlights the potential of using a digital pipeline to assist tau aggregate type-specific quantification in different brain regions,” said Tanrada Pansuwan, PhD, the first and corresponding author of this study. “This could be adapted to assist postmortem diagnosis in a clinical setting or be used for research purposes to better understand the progression of tau pathology in PSP.”

The report was published in Acta Neuropathologica Communications.

 

Rationale: Improving Tau Quantification

PSP is a neurodegenerative disease characterized by the accumulation of insoluble tau protein deposits, which cause damage to critical cortical and subcortical regions of the brain.1 The current standard method for examining postmortem tau histology involves a semiquantitative approach in which pathologists visually examine postmortem brain sections and grade the severity of tau pathology as absent, mild, moderate, or severe. However, this approach is time consuming, requires extensive training, and is prone to inter-rater variability.1

The authors aimed to address these limitations by developing a digital pathology pipeline using a probabilistic random forest algorithm to quantify neuronal and glial tau densities in PSP.

“We chose PSP as our first candidate disease because neuronal and glial tau aggregates in PSP are relatively distinct, and PSP has a well-defined postmortem staging scheme for pipeline validation,” Dr. Pansuwan explained.

 

Approach

By employing a sophisticated random forest machine learning classifier, researchers developed a versatile pipeline to analyze tau deposits in various PSP brain regions.1Tailored classifiers were developed to quantify four tau aggregate types: neurofibrillary tangles, coiled bodies, tufted astrocytes, and tau fragments. The model was designed to analyze tau aggregates in several regions of the brain, including cortical and subcortical regions.

“In designing the pipeline to be applicable across brain regions, we considered that there are not equal numbers of neuronal and glial cells in the brain, leading to a class imbalance of tau aggregates for the machine learning model,” said Dr. Pansuwan. Class imbalance and tau morphology also differ across brain regions, leading to inherent ambiguity in classifying tau aggregates.

“With these challenges in mind, we used a balanced random forest classifier to tackle class imbalance, group brain regions with similar characteristics together, and develop a classifier specific to each brain grouping,” she added.

In addition to incorporating built-in balanced sampling and thresholding steps into the model architecture, the team also added measures to exclude objects for which the classifier is uncertain.1 This aimed to address the inherent ambiguity in the classification of tau aggregates.

Commenting on the novelty of this approach, Dr. Pansuwan said: “This is the first study to have developed and tested a digital tau pathology pipeline on a large number of brain regions, including both cortical, basal ganglia nuclei and dentate nucleus, specifically in PSP.”

 

The Machine Learning Pipeline Accurately Classifies Tau Pathologies

In a proof-of-concept study, the researchers validated the machine learning pipeline using 227 brain slides from 32 patients with PSP. The pipeline demonstrated high classification performance (F1 scores > 0.90), comparable to expert neuropathologist assessment across multiple cortical and subcortical brain regions, including the cortex, basal ganglia, and cerebellum.1

Moreover, the pipeline provided more sensitive and granular quantification than standard semiquantitative methods. When validated against the current consensus PSP staging criteria, cortical tufted astrocyte density and subcortical coiled body density showed the highest correlation with the overall PSP stage (r = 0.62 and r = 0.38, respectively).1

“The finding that tau burden quantified using our pipeline correlated with the current consensus PSP postmortem staging scheme demonstrates the reliability of the digital pipeline,” noted Dr. Pansuwan.

The study also showed that digitally quantified cortical tau density and neurofibrillary tangle density in subcortical regions were correlated with clinical severity measured by the PSP rating scale score.1

“Our data suggest the importance of studying tau aggregate type-specific burden in different brain regions, as opposed to overall tau, to gain insights into the pathogenesis and progression of tauopathies,” she added.

 

Future Directions

The findings of this proof-of-concept study suggest that the automated quantitative digital pathology pipeline enables the analysis of tau type, distribution, and severity at scale to elucidate the mechanisms of disease progression in tauopathies. However, the limited sample size hindered the analysis of a broader spectrum of PSP subtypes and common co-pathologies.

“The pipeline remains to be tested on other PSP cohorts to confirm its generalizability beyond the Cambridge Brain Bank cohort. It also remains to be determined whether the pipeline is sensitive enough to detect differences in tau distribution in the brains of individuals with different PSP subtypes,” said Dr. Pansuwan.

Further studies are needed to automate the removal of confounding non-tau deposits, such as iron granules. Incorporating additional affected regions, such as the midbrain, would also enhance neuropathological coverage and insights. Furthermore, future efforts are needed to extend this pipeline to other tauopathies and neurodegenerative proteinopathies.

Although improvements are needed, this study represents a key step toward the use of automated methods to quantify tau type-specific burden in different brain regions and correlate it with clinical severity.

 

The study was funded by the NIHR Cambridge Biomedical Research Centre, Wellcome Trust, Medical Research Council, PSP Association, and Lundbeck Foundation.

 

References

  1. Pansuwan T, Quaegebeur A, Kaalund SS, et al. Accurate digital quantification of tau pathology in progressive supranuclear palsy. Acta Neuropathol Commun. 2023;11(1):178. Published 2023 Nov 9. doi:10.1186/s40478-023-01674-y

Share This Post

Leave a Reply