Consistent and effective method to define the mouse estrous cycle stage by a deep learning-based model.

Area of Science:

• Veterinary Science
• Computational Biology
• Reproductive Biology

Background:

• The mouse estrous cycle influences reproductive hormone levels and experimental outcomes.
• Accurate staging of the estrous cycle is crucial for reliable female mouse research.
• Traditional microscopic analysis of vaginal smears is time-consuming and subjective.

Purpose of the Study:

• To develop an accurate and reproducible method for staging the mouse estrous cycle using digital whole-slide images (WSIs).
• To implement a deep convolutional neural network (CNN) for automated estrous cycle staging.
• To validate the AI model's performance against human expert evaluation.

Main Methods:

• Developed a CNN model on the Aiforia Create platform for semantic segmentation of vaginal smear WSIs.
• Trained the CNN using 171 hematoxylin-stained samples with pixel-level annotations.
• Validated the model against 148 slides from three independent studies, comparing CNN results with four researchers.

Main Results:

• The CNN model demonstrated excellent agreement (Fleiss kappa = 0.75) with human evaluators for overall estrous cycle staging.
• High kappa values were achieved for diestrus (0.89), estrus (0.79), and proestrus (0.74).
• Metestrus staging showed lower agreement (kappa = 0.26) due to its short duration and ambiguous definition.

Conclusions:

• The developed CNN model provides a reliable and effective tool for classifying mouse estrous cycle stages from digital vaginal smears.
• This AI-driven approach offers a reproducible alternative to manual microscopic evaluation.
• Further refinement may improve metestrus stage identification, enhancing the model's utility in research.