Quantification of Hepatocellular Mitoses in a Toxicological Study in Rats Using a Convolutional Neural Network

Area of Science:

• Toxicologic Histopathology
• Computational Pathology
• Biomarker Discovery

Background:

• Convolutional neural networks (CNNs) are increasingly utilized for quantitative analysis in histopathology.
• Assessing hepatocyte cell proliferation is crucial in toxicity studies.
• Traditional methods for proliferation assessment can be time-consuming and subjective.

Purpose of the Study:

• To evaluate the performance of a CNN-based (Halo-AI) system for mitotic figure detection in hepatocytes.
• To compare CNN-based detection with pathologist assessment and immunohistochemistry labeling indices (LIs) for Ki-67 and 5-bromodeoxyuridine (BrdU).
• To explore the correlation between different proliferation assessment methods.

Main Methods:

• Utilized CNN-based (Halo-AI) mitotic figure detection on hepatocyte tissues from a GSK-3 inhibitor toxicity study.
• Compared CNN results with manual pathologist detection.
• Compared CNN results with Ki-67 and BrdU immunohistochemistry labeling indices (LIs) analyzed by image analysis.

Main Results:

• CNN-based mitotic figure detection was accurate and faster than pathologist assessment.
• CNN results showed comparability to Ki-67 and BrdU LIs.
• Moderate correlation was observed between different proliferation assessment methods, potentially due to cell cycle differences and test item pharmacology.

Conclusions:

• CNNs provide an efficient and standardized method for hepatocyte cell proliferation assessment in toxicologic histopathology.
• CNNs can reduce pathologist workload and improve result standardization, adhering to the 3R principles for animal welfare.
• The choice of proliferation assessment method should consider the specific cell cycle components captured and potential drug effects.