# Advanced Multiscale Attention Network for Estrous Cycle Stage Identification from Rat Vaginal Cytology

## TL;DR

This paper introduces SLENet, a deep learning model that accurately identifies rat estrous cycle stages from vaginal smear images, improving efficiency and reducing human error.

## Contribution

The novel Spatial Efficient Channel Attention mechanism and non-local attention in SLENet achieve higher accuracy than existing models for estrous cycle classification.

## Key findings

- SLENet achieved 96.31% accuracy on a dataset of 2655 rat vaginal smear images.
- The model outperformed EfficientNet by 2.11% in classification accuracy.

## Abstract

Understanding the estrous cycle of female rats is crucial for ensuring the reliability of biomedical experiments, as hormonal fluctuations can significantly affect drug responses and physiological behaviors. However, traditional manual identification of estrous stages from microscopic images is time-consuming, subjective, and requires specialized expertise. In this study, we developed a deep learning model called SLENet to automatically classify the four stages of the rat estrous cycle (proestrus, estrus, metestrus, diestrus) based on vaginal smear images. By introducing spatial and global attention mechanisms, our model achieved a high accuracy of 96.31 on a curated dataset of 2655 images. This approach not only improves classification performance compared to existing models but also reduces human workload, providing a reliable tool for researchers in reproductive biology and pharmacological studies.

In clinical medicine, rats are commonly used as experimental subjects. However, their estrous cycle significantly impacts their biological responses, leading to differences in experimental results. Therefore, accurately determining the estrous cycle is crucial for minimizing interference. Manually identifying the estrous cycle in rats presents several challenges, including high costs, long training periods, and subjectivity. To address these issues, this paper proposes a classification network, Spatial Long-distance EfficientNet (SLENet). This network is designed based on EfficientNet, specifically modifying the Mobile Inverted Bottleneck Convolution (MBConv) module by introducing a novel Spatial Efficient Channel Attention (SECA) mechanism to replace the original Squeeze Excitation (SE) module. Additionally, a non-local attention mechanism is incorporated after the last convolutional layer to enhance the network’s ability to capture long-range dependencies. On 2655 microscopy images of rat vaginal epithelial cells (with 531 test), SLENet achieves 96.31% accuracy, surpassing EfficientNet (94.20%). This finding provides practical value for optimizing experimental design in rat-based studies such as reproductive and pharmacological research, but this study is limited to microscopy image data, without considering other factors; thus, future work could incorporate temporal pattern and multi-modal inputs to further enhance robustness.