Imaging myelin degradation in ex vivo prefrontal cortex tissue blocks in Alzheimer's disease and chronic traumatic encephalopathy

Abstract

INTRODUCTIONAlzheimer's disease (AD) and chronic traumatic encephalopathy (CTE) are tauopathies with gray matter (GM) myelin changes that are challenging to assess with standard imaging. New methods are needed to quantify myelin integrity in autopsy brain tissues.METHODSWe used polarization‐sensitive optical coherence tomography (PS‐OCT) to measure bulk tissue relative retardance and birefringence microscopy for high‐resolution imaging of myelin degradation. Samples included five AD, five CTE, and four age‐matched normal controls.RESULTSWhen controlling for age and postmortem interval, no statistically significant differences in white matter retardance or GM myelin defect density were observed between groups. The age difference between controls (64 ± 4.7 years, mean ± SD) and disease groups (80.3 ± 7 years) emerged as an important confounding factor. Amyloid beta and tau staining showed weak correlations with myelin defects.DISCUSSIONOur label‐free approach enables large‐volume imaging of brain tissue, a valuable tool for studying myelin changes in neurodegenerative diseases.Highlights Multi‐modal assessment of myelin integrity using polarization‐sensitive optical coherence tomography (PS‐OCT) and high‐resolution birefringence microscopy. Age emerged as a critical confounding factor; no significant disease differences were found. Weak correlation between myelin defects and deposition of amyloid beta/tau was found in prefrontal gray matter. Label‐free optical methods enable high‐resolution, large‐volume imaging of myelin.

Accelerating myelin defect detection in neurodegenerative disorders: a human-in-the-loop deep learning approach with birefringence microscopy

“…The model trained on RGB data converted to grayscale format (Sec. 2.4.2 ) encountered an unexpected domain shift when applied to authentic grayscale CCP-BRM images from our previous study, 9 which resulted in increased false positive rates due to transverse axons and blood vessels being misclassified as vesicles. We addressed this by performing three additional iterations of our training workflow with additional brain tissue images acquired in authentic grayscale CCP-BRM format for this domain adaptation purpose: (1) generating predictions on unlabeled images, (2) having experts identify correct predictions and label false positives, and (3) retraining the model with the newly added expert-verified data.…”

“…The samples were cut into -thick sections ( area) using a vibratome, wet-mounted, and index-matched in 85% glycerol (to minimize scattering), following established protocols. 16 These tissue samples and the associated expert consensus annotations were previously utilized in our validation study of manual CCP-BRM analysis, 9 enabling direct comparison between manual and automated approaches on identical datasets.…”

“…CCP-BRM imaging was performed using a commercial slide-scanner microscope (Olympus VS-120) modified with circular polarizers. 9 The RGB CCP-BRM implementation utilized “white” light illumination (XCite 120LED light engine) and a color camera (AVT Pike F-505C), providing real-time fiber orientation encoding with operational advantages over traditional quantitative birefringence microscopy (qBRM) 16 approaches ( Fig. 2 demonstrates these comparative advantages).…”

“…Annotations were categorized into two classes: general myelin “defects” (including breaks, delaminations, blebbings, swellings, tears, and irregularities) and spherical “lipid vesicles” formed from myelin debris. These defect categories and tissue types were established in our previous comprehensive characterization study, 9 ensuring consistency with established pathological criteria across control, AD, and CTE conditions. Such defects can impair saltatory conduction of action potentials and contribute to functional decline in neurodegenerative conditions.…”

“…Such defects can impair saltatory conduction of action potentials and contribute to functional decline in neurodegenerative conditions. 3 Expert consensus annotations were also established following protocols from our previous validation study, 9 which included expert neuropathologists as co-authors, and where three trained annotators achieved good inter-rater reliability through standardized defect identification criteria.…”

Accelerating myelin defect detection in neurodegenerative disorders: a human-in-the-loop deep learning approach with birefringence microscopy

“…The model trained on RGB data converted to grayscale format (Sec. 2.4.2 ) encountered an unexpected domain shift when applied to authentic grayscale CCP-BRM images from our previous study, 9 which resulted in increased false positive rates due to transverse axons and blood vessels being misclassified as vesicles. We addressed this by performing three additional iterations of our training workflow with additional brain tissue images acquired in authentic grayscale CCP-BRM format for this domain adaptation purpose: (1) generating predictions on unlabeled images, (2) having experts identify correct predictions and label false positives, and (3) retraining the model with the newly added expert-verified data.…”

“…The samples were cut into -thick sections ( area) using a vibratome, wet-mounted, and index-matched in 85% glycerol (to minimize scattering), following established protocols. 16 These tissue samples and the associated expert consensus annotations were previously utilized in our validation study of manual CCP-BRM analysis, 9 enabling direct comparison between manual and automated approaches on identical datasets.…”

“…CCP-BRM imaging was performed using a commercial slide-scanner microscope (Olympus VS-120) modified with circular polarizers. 9 The RGB CCP-BRM implementation utilized “white” light illumination (XCite 120LED light engine) and a color camera (AVT Pike F-505C), providing real-time fiber orientation encoding with operational advantages over traditional quantitative birefringence microscopy (qBRM) 16 approaches ( Fig. 2 demonstrates these comparative advantages).…”

“…Annotations were categorized into two classes: general myelin “defects” (including breaks, delaminations, blebbings, swellings, tears, and irregularities) and spherical “lipid vesicles” formed from myelin debris. These defect categories and tissue types were established in our previous comprehensive characterization study, 9 ensuring consistency with established pathological criteria across control, AD, and CTE conditions. Such defects can impair saltatory conduction of action potentials and contribute to functional decline in neurodegenerative conditions.…”

“…Such defects can impair saltatory conduction of action potentials and contribute to functional decline in neurodegenerative conditions. 3 Expert consensus annotations were also established following protocols from our previous validation study, 9 which included expert neuropathologists as co-authors, and where three trained annotators achieved good inter-rater reliability through standardized defect identification criteria.…”

Accelerating myelin defect detection in neurodegenerative disorders: a human-in-the-loop deep learning approach with birefringence microscopy

“…The model trained on RGB data converted to grayscale format (Sec. 2.4.2 ) encountered an unexpected domain shift when applied to authentic grayscale CCP-BRM images from our previous study, 9 which resulted in increased false positive rates due to transverse axons and blood vessels being misclassified as vesicles. We addressed this by performing three additional iterations of our training workflow with additional brain tissue images acquired in authentic grayscale CCP-BRM format for this domain adaptation purpose: (1) generating predictions on unlabeled images, (2) having experts identify correct predictions and label false positives, and (3) retraining the model with the newly added expert-verified data.…”

“…The samples were cut into -thick sections ( area) using a vibratome, wet-mounted, and index-matched in 85% glycerol (to minimize scattering), following established protocols. 16 These tissue samples and the associated expert consensus annotations were previously utilized in our validation study of manual CCP-BRM analysis, 9 enabling direct comparison between manual and automated approaches on identical datasets.…”

“…CCP-BRM imaging was performed using a commercial slide-scanner microscope (Olympus VS-120) modified with circular polarizers. 9 The RGB CCP-BRM implementation utilized “white” light illumination (XCite 120LED light engine) and a color camera (AVT Pike F-505C), providing real-time fiber orientation encoding with operational advantages over traditional quantitative birefringence microscopy (qBRM) 16 approaches ( Fig. 2 demonstrates these comparative advantages).…”

“…Annotations were categorized into two classes: general myelin “defects” (including breaks, delaminations, blebbings, swellings, tears, and irregularities) and spherical “lipid vesicles” formed from myelin debris. These defect categories and tissue types were established in our previous comprehensive characterization study, 9 ensuring consistency with established pathological criteria across control, AD, and CTE conditions. Such defects can impair saltatory conduction of action potentials and contribute to functional decline in neurodegenerative conditions.…”

“…Such defects can impair saltatory conduction of action potentials and contribute to functional decline in neurodegenerative conditions. 3 Expert consensus annotations were also established following protocols from our previous validation study, 9 which included expert neuropathologists as co-authors, and where three trained annotators achieved good inter-rater reliability through standardized defect identification criteria.…”

Accelerating myelin defect detection in neurodegenerative disorders: a human-in-the-loop deep learning approach with birefringence microscopy

“…The model trained on RGB data converted to grayscale format (Sec. 2.4.2 ) encountered an unexpected domain shift when applied to authentic grayscale CCP-BRM images from our previous study, 9 which resulted in increased false positive rates due to transverse axons and blood vessels being misclassified as vesicles. We addressed this by performing three additional iterations of our training workflow with additional brain tissue images acquired in authentic grayscale CCP-BRM format for this domain adaptation purpose: (1) generating predictions on unlabeled images, (2) having experts identify correct predictions and label false positives, and (3) retraining the model with the newly added expert-verified data.…”

“…The samples were cut into -thick sections ( area) using a vibratome, wet-mounted, and index-matched in 85% glycerol (to minimize scattering), following established protocols. 16 These tissue samples and the associated expert consensus annotations were previously utilized in our validation study of manual CCP-BRM analysis, 9 enabling direct comparison between manual and automated approaches on identical datasets.…”

“…CCP-BRM imaging was performed using a commercial slide-scanner microscope (Olympus VS-120) modified with circular polarizers. 9 The RGB CCP-BRM implementation utilized “white” light illumination (XCite 120LED light engine) and a color camera (AVT Pike F-505C), providing real-time fiber orientation encoding with operational advantages over traditional quantitative birefringence microscopy (qBRM) 16 approaches ( Fig. 2 demonstrates these comparative advantages).…”

“…Annotations were categorized into two classes: general myelin “defects” (including breaks, delaminations, blebbings, swellings, tears, and irregularities) and spherical “lipid vesicles” formed from myelin debris. These defect categories and tissue types were established in our previous comprehensive characterization study, 9 ensuring consistency with established pathological criteria across control, AD, and CTE conditions. Such defects can impair saltatory conduction of action potentials and contribute to functional decline in neurodegenerative conditions.…”

“…Such defects can impair saltatory conduction of action potentials and contribute to functional decline in neurodegenerative conditions. 3 Expert consensus annotations were also established following protocols from our previous validation study, 9 which included expert neuropathologists as co-authors, and where three trained annotators achieved good inter-rater reliability through standardized defect identification criteria.…”