Machine learning assisted neuropathological assessment of 126 brains from PSEN1 E280A familial Alzheimer's disease. (20th December 2022)

• Record Type: Journal Article
• Authors: Villalba‐Moreno, Nelson David; Littau, Tim; Littau, Jessica Lisa; Mejia‐Cupajita, Juan Pablo; Hartmann, Kristin; Cardona‐Madrigal, Duvan; Krasemann, Susanne; Glatzel, Markus; Lopera, Francisco; Villegas‐Lanau, Andres; Sepulveda‐Falla, Diego
• Is Part Of: Alzheimer's & dementia. Volume 18(2022)Supplement 4
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-12-20
• Subjects: Alzheimer's disease -- Periodicals; Alzheimer Disease -- Periodicals; Dementia -- Periodicals; Démence; Maladie d'Alzheimer; Périodique électronique (Descripteur de forme); Ressource Internet (Descripteur de forme); 616.83
• DOI: 10.1002/alz.069409
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: During the last 25 years we have collected 126 brains from the largest family in the world carrying PSEN1 mutation E280A and suffering from familial Alzheimer's disease (FAD). Classical neuropathology scales rate all FAD cases as severe, overlooking the degree of heterogeneity observed in these cases. Here we present a new approach involving High Throughput Digital Neuropathology (HTDNp), Machine Learning (ML) classification of pathology, and multidimensional data analysis. Method: Frontal, Temporal, Parietal, and Occipital cortices, and cerebellum of all 126 brains were processed for H&E staining and immunohistochemistry (IHC) for amyloid beta (Ab) and hyperphosphorylated tau (pTau). Cases were diagnosed, and staged according to CERAD, Braak and Thal systems. All slides were digitally scanned and processed for ML classification. Ab pathology was classified as core and diffuse plaques, and cerebral amyloid angiopathy (CAA), using a previously trained convolutional neural network (CNN). pTau pathology was classified as neurofibrillary tangles, dystrophic neurites, and diffuse pathology, by training a new CNN. ML was used for data dimensionality reduction, using self‐organizing maps (SOM). Clinical profiles were identified in the pathological clusters, analyzing presence and onset of specific symptoms. Result: PSEN1 E280A cases showed high degree of heterogeneity regarding ages of onset, disease duration, brain weight, and ApoE haplotype. Clinically, memory and language impairment were prevalent, while depression, insomnia, gait disorders and behavioral symptoms presented only in some cases. All cases were classified as CERAD C, Thal 5, and Braak VI. ML‐assisted classification and quantification of Ab and pTau pathology showed a high degree of variability among patients and brain areas. SOM analysis identified four distinct pathological clusters, each one with a characteristic clinical profile, indicating a clinical correlate for FAD pathological presentation in these cases. Conclusion: Current neuropathological diagnosis of Alzheimer's disease is not sensitive enough to differentiate among disease subtypes or heterogeneity. This can be detected by using HTDNp and ML tools, allowing the identification of clinically relevant pathological subtypes even in PSEN1 E280A brains, an otherwise uniform sample. This approach can be explored further in other FAD and sporadic Alzheimer's disease cases, aiming for a more precise diagnosis, and a better understanding of the disease.