Automated identification of multinucleated germ cells with U-Net

Area of Science:

• Reproductive Toxicology
• Computational Pathology
• Developmental Biology

Background:

• Phthalic acid esters (phthalates) are known male reproductive toxicants.
• Phthalate exposure during fetal development can lead to abnormal multinucleated germ cells (MNGs).
• Identifying MNGs in histological sections is time-consuming and requires specialized expertise, hindering routine quantification in toxicity studies.

Purpose of the Study:

• To develop an automated method for detecting MNGs in histological images.
• To standardize and accelerate the quantification of MNGs in phthalate toxicity assessments.
• To improve the efficiency of evaluating male reproductive toxicants.

Main Methods:

• A convolutional neural network with a U-Net architecture was trained using hand-labeled histological images containing MNGs.
• The model was trained to identify MNGs in unlabeled histological images.
• Model performance was evaluated using unseen hand-labeled images across five data configurations.

Main Results:

• The automated detection model achieved performance comparable to human identification.
• In the best configuration, the model's accuracy exceeded that of human identification.
• The developed method significantly speeds up the identification and quantification of MNGs.

Conclusions:

• Automated image analysis provides a rapid and standardized approach for MNG detection.
• This method facilitates more extensive data collection for phthalate toxicity studies.
• The tool enables more efficient assessment of MNGs as a histopathological endpoint.