Deep learning-based image-analysis algorithm for classification and quantification of multiple histopathological...

Area of Science:

• Preclinical toxicology
• Digital pathology
• Artificial intelligence in drug discovery

Background:

• AI-based image analysis is gaining traction in pharmaceutical preclinical safety assessment.
• Histopathological evaluation of whole slide images (WSIs) is crucial but can be time-consuming.

Purpose of the Study:

• To develop and validate an AI-based solution for detecting, classifying, and quantifying histopathological findings in rat liver WSIs.
• To assess the performance of AI algorithms in supporting toxicological studies.

Main Methods:

• A U-Net-based deep learning network was trained to identify seven common histopathological findings in rat liver WSIs.
• Algorithms were validated on 255 liver WSIs, assessing detection, classification, and quantification accuracy.

Main Results:

• AI algorithms demonstrated good performance in detecting abnormal areas, correctly classifying approximately 75% of specimens.
• Findings with clear boundaries, like vacuolation and single-cell necrosis, were accurately detected.
• Quantitative analyses correlated well with pathologist diagnoses, though performance varied for findings with ambiguous boundaries (e.g., hepatocellular hypertrophy).

Conclusions:

• Deep learning algorithms can simultaneously detect, classify, and quantify multiple histopathological findings in rat liver WSIs.
• AI serves as a valuable supportive tool for histopathological evaluation, particularly for primary screening in rat toxicity studies.