Urli Susy, Corte Pause Francesca, Dreossi Talissa, Crociati Martina, Stradaioli Giuseppe
Department of Agricultural, Food, Environmental and Animal Sciences, University of Udine, Via Delle Scienze 206, 33100, Udine, Italy.
Department of Mathematics, Computer Science and Physics, University of Udine, Via Delle Scienze 206, 33100, Udine, Italy; GNCS-INdAM, Gruppo Nazionale per Il Calcolo Scientifico, Istituto Nazionale di Alta Matematica, Città Universitaria, P.le Aldo Moro, 5, 00185, Roma, Italy.
Theriogenology. 2025 Oct 1;245:117504. doi: 10.1016/j.theriogenology.2025.117504. Epub 2025 May 29.
The morphological characteristics of bull spermatozoa are usually evaluated visually using bright-field microscopy according to the guidelines proposed by the Society for Theriogenology (SFT) for the Bull Breeding Soundness Evaluation (BBSE). However, analysis is labor consuming and requires experienced personnel to obtain reliable results. Nevertheless, the artificial insemination industry increasingly demands the implementation of genomic selection schemes for young bulls. Hence, there is a growing need for a more standardized technique to analyze semen quality, particularly for the evaluation of spermatozoa abnormalities that affect semen freezing suitability and fertilizing capacity. Therefore, an Artificial Intelligence (AI) algorithm for the automated classification of microscope-acquired images of spermatozoa was developed using neural networks, specifically YOLO networks, based on convolutional neural networks (CNNs) that were able to learn and extract relevant features from complex visual data through image segmentation. The aim was to assess the algorithm ability to identify sperm cells in microscope-acquired images, establish their viability and to classify morphology based on a simplified scheme which included only normal or major/minor defect categories. The dataset comprised 8243 images, which were labeled and annotated with bounding boxes to allow the segmentation algorithm to learn. The performance obtained by the algorithm showed an accuracy of 82 %, although it was not observed for all classes (excluding a probable case of overfitting where accuracy reached 100 %), and a precision of 85 % in the correct classification of spermatozoa morphology. Results thereby confirmed the potential applicability of the algorithm in bull semen analysis without excluding its future implementation for achieving optimal performance.
通常根据家畜繁殖学学会（SFT）提出的公牛繁殖健全性评估（BBSE）指南，使用明场显微镜对公牛精子的形态特征进行目视评估。然而，这种分析工作繁琐，需要有经验的人员才能获得可靠的结果。尽管如此，人工授精行业对实施青年公牛基因组选择方案的需求日益增加。因此，越来越需要一种更标准化的技术来分析精液质量，特别是用于评估影响精液冷冻适用性和受精能力的精子异常情况。因此，基于卷积神经网络（CNN）开发了一种人工智能（AI）算法，用于对显微镜采集的精子图像进行自动分类，具体是使用基于能够通过图像分割从复杂视觉数据中学习和提取相关特征的YOLO网络的神经网络。目的是评估该算法在显微镜采集图像中识别精子细胞、确定其活力并根据仅包括正常或主要/次要缺陷类别的简化方案对形态进行分类的能力。该数据集包含8243张图像，这些图像用边界框进行了标记和注释，以便分割算法学习。该算法获得的性能显示准确率为82%，尽管并非所有类别都达到了这一准确率（不包括可能的过拟合情况，即准确率达到100%），并且在精子形态的正确分类中精度为85%。结果证实了该算法在公牛精液分析中的潜在适用性，同时并不排除其未来为实现最佳性能而进行的实施。
