VISEM-Tracking, a human spermatozoa tracking dataset.

Area of Science:

• Reproductive Biology
• Computer Vision
• Machine Learning

Background:

• Manual sperm motility assessment via microscopy is challenging due to rapid spermatozoan movement and requires extensive expert training.
• Computer-aided sperm analysis (CASA) is increasingly adopted in clinical settings, but requires more data for supervised machine learning to enhance accuracy.
• Existing datasets lack sufficient annotated data for robust machine learning model training in sperm analysis.

Purpose of the Study:

• To introduce the VISEM-Tracking dataset, a novel resource for advancing sperm motility and kinematics analysis.
• To provide a comprehensive dataset including manually annotated sperm characteristics and unlabeled video clips.
• To establish baseline performance metrics for sperm detection using deep learning models on the new dataset.

Main Methods:

• The VISEM-Tracking dataset comprises 20 video recordings (30 seconds each, 29,196 frames) of wet semen preparations.
• Data includes manually annotated bounding-box coordinates and expert-analyzed sperm characteristics.
• Baseline sperm detection performance was evaluated using the YOLOv5 deep learning model trained on the VISEM-Tracking dataset.

Main Results:

• The VISEM-Tracking dataset facilitates the training of complex deep learning models for spermatozoa analysis.
• Baseline YOLOv5 model achieved promising sperm detection performance, demonstrating the dataset's utility.
• The dataset's annotated and unlabeled data support various machine learning approaches, including self- and unsupervised learning.

Conclusions:

• The VISEM-Tracking dataset is a valuable resource for improving automated sperm analysis.
• The dataset enables the development of more accurate and reliable computer-aided sperm analysis tools.
• This work highlights the potential of deep learning models trained on comprehensive datasets for advancing reproductive health diagnostics.