A Deep Learning-Based Computational Framework for the analysis of neurofibrillary tangles in post-mortem brain micrographs from alzheimer’s patients using object detection and semi-automatic segmentation

Neurofibrillary tangles (NFTs) are a pathological hallmark of Alzheimer’s disease (AD) and related tauopathies, consisting of abnormal accumulations of the tau protein. Immunofluorescence microscopy remains the standard method for visualizing these aggregates, yet its manual interpretation is time-consuming and prone to variability. Their precise quantification is crucial for understanding disease progression, as it allows researchers to correlate NFT burden with cognitive decline, providing valuable insights into the underlying mechanisms of neurodegeneration. However, the labor-intensive nature of manual assessment and its susceptibility to observer variability limit scalability, highlighting the need for automated, reproducible methodologies in large-scale studies. To address these limitations, we present a deep learning-based computational framework for automated detection, segmentation, and quantitative analysis of NFTs in post-mortem brain micrographs from AD patients. Our approach integrates state-of-the-art object detectors—YOLO11/v12, Faster R-CNN, and transformerbased DETR/RT-DETR—with the Segment Anything Model (SAM) to refine bounding boxes into pixel-accurate masks. Evaluated on a curated dataset of over 900 hippocampal and entorhinal micrographs, our framework achieves an mAP50 of 0.81 for detection and a mean IoU of 0.86 for segmentation. Additionally, we conduct a comprehensive NFT burden analysis across brain regions, highlighting the hippocampal subiculum as the most affected area. These results demonstrate the potential of deep learning to enable high-throughput and reproducible NFT quantification, supporting large-scale neuropathological studies.

https://orcid.org/0000-0001-5597-939X

54789443800