Automated Computational Pathology to Assess Degenerative Disc Histology.

Area of Science:

• Biomedical Engineering
• Computational Pathology
• Histology

Background:

• Preclinical models are crucial for understanding disc degeneration pathology.
• Traditional histopathology scoring is time-consuming and relies on manual grading.
• Computational pathology offers a more efficient approach using machine learning (ML).

Purpose of the Study:

• To develop a segmentation model for identifying seven distinct disc tissues.
• To utilize ML-derived measures for estimating pathological changes in disc degeneration.
• To correlate computational measures with traditional histological scoring.

Main Methods:

• Developed a deep convolutional neural network with active learning for tissue segmentation.
• Applied the model to Hematoxylin and eosin-stained rat disc sections from four studies.
• Extracted geometric and cell count features to correlate with standard histologic scores.

Main Results:

• The segmentation model achieved high performance (mIOU 0.83, fwIOU 0.94).
• ML-derived measures strongly correlated with histologic scores (rho = 0.65–0.87).
• Significant alterations in NP cell number, area ratio, NP/AF border, NP roundness, and AF perimeter were detected.

Conclusions:

• Developed the first computational pathology model for disc degeneration phenotyping.
• The model correlates with traditional histopathology and detects subtle differences.
• Automated analysis increases efficiency and measures features beyond standard scoring.