Development of a deep learning-based histological evaluation model for critical-size bone defect healing in rats - an...

Area of Science:

• Biomedical Engineering
• Regenerative Medicine
• Computational Pathology

Background:

• Critical-size femoral defects in rats are a standard preclinical model for bone regeneration.
• Histological evaluation is crucial but time-consuming and prone to observer variability.
• Machine learning, specifically CNNs, presents an opportunity for objective histological analysis.

Purpose of the Study:

• To develop and validate a CNN model for automated histological assessment of bone healing.
• To compare the AI's performance against human experts and students.

Main Methods:

• A modified U-Net CNN was trained on Movat pentachrome-stained rat femoral defect histology (n=669).
• Five tissue classes were annotated for semantic segmentation and classification of healing stages.
• The AI's bone healing scores were compared to expert and student assessments on 20 independent images.

Main Results:

• The CNN model demonstrated high segmentation accuracy for key tissue types.
• AI-generated healing scores strongly correlated with expert ratings (Spearman r=0.819).
• AI performance in scoring was comparable to medical students and showed excellent agreement with experts (ICC=0.820).

Conclusions:

• A CNN-based model achieves expert-level performance in histological bone healing assessment.
• This AI tool provides a reproducible and time-efficient solution for preclinical bone regeneration research.