Ovarian Toxicity Assessment in Histopathological Images Using Deep Learning

Area of Science:

• Toxicologic Pathology
• Digital Pathology
• Computational Biology

Background:

• Ovarian toxicity is a critical safety concern in preclinical cancer therapeutic development.
• Accurate identification of ovarian pathology is essential for drugs targeting women of child-bearing potential.
• Traditional histopathological assessment using hematoxylin and eosin (H&E) staining is the current standard but can be labor-intensive.

Purpose of the Study:

• To develop an automated deep learning algorithm for rapid and accurate enumeration of rat ovarian corpora lutea.
• To validate the algorithm's performance against pathologist-based assessment on H&E-stained slides.
• To assess the algorithm's utility in identifying reproductive toxicity trends in preclinical studies.

Main Methods:

• Development of a deep learning network algorithm for automated counting of corpora lutea.
• Testing the algorithm on H&E-stained ovarian tissues from five rat toxicity studies.
• Comparative analysis of algorithm counts versus pathologist evaluations.

Main Results:

• The deep learning algorithm accurately enumerated corpora lutea across all tested studies.
• The method demonstrated comparable accuracy to manual pathologist assessment.
• The algorithm successfully identified distinct trends in studies with and without observed reproductive toxicity.

Conclusions:

• Automated deep learning offers a rapid and accurate method for quantifying rat ovarian corpora lutea.
• This digital pathology approach can significantly aid toxicologic pathology assessments in preclinical drug development.
• The tool has the potential to enhance the efficiency and reliability of reproductive toxicity evaluations.