TDP-43 Inclusion Detection (YOLOv11)

Automated detection of TDP-43 inclusions in histopathological images for frontotemporal dementia (FTD), amyotrophic lateral sclerosis (ALS) and Limbic-predominant Age-related TDP-43 Encephalopathy (LATE)

Performance

• mAP: 0.520
• Accuracy: 94.2% (highly specific)
• Dataset: 9 WSI with ~450 expert annotations

Quick Start

from ultralytics import YOLO
from huggingface_hub import hf_hub_download

# Download and load model
model_path = hf_hub_download(
    repo_id="Center-for-Computational-Neuropathology/TDP-43",
    filename="best.pt"
)
model = YOLO(model_path)

# Run inference (use higher confidence for clinical use)
results = model.predict("tdp43_stained_image.jpg", conf=0.4, imgsz=640)

Clinical Relevance

Detects TDP-43 inclusions in:

• Frontotemporal Dementia (FTD)
• Amyotrophic Lateral Sclerosis (ALS)
• FTLD-TDP subtypes
• Limbic-predominant Age-related TDP-43 Encephalopathy (LATE)

Key Features

✅ 94.2% accuracy - High specificity, low false positives
✅ Conservative strategy ideal for clinical screening
✅ Reliable when detection is made
⚠️ May miss subtle or atypical inclusions

Training Insights

This model required early termination due to validation instabilities, highlighting the unique computational challenges of TDP-43 morphology compared to other pathologies (e.g., Lewy bodies achieved stable 196-epoch training).

Limitations

• Conservative approach may miss subtle inclusions
• Requires TDP-43 immunohistochemistry (phospho-TDP-43)
• Cannot distinguish TDP-43 types (A, B, C, etc.)
• Requires expert validation for clinical use

Citation

@article{neuropath_yolo_2025,
  title={Automated Detection of Neurodegenerative Pathology Using YOLOv11},
  author={[Authors]},
  journal={[Journal]},
  year={2025}
}