Joint Detection and Tracking of Sperm for Improving Motility Analysis
Nordic Machine Intelligence · 29.04.2026 · 6 (3) · 30-34
DOI: 10.5617/nmi.13365 ↗Kurz gefasst
Fachartikel in der Zeitschrift Nordic Machine Intelligence, online erschienen am 29.04.2026. Verfasst von Saher Tariq und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Male fertility assessment plays a critical role in the diagnosis of human infertility, where semen quality is primarily evaluated based on two key factors: motility and morphology. In existing eHealth solutions, such as Computer-Assisted Sperm Analysis (CASA), motility analysis is often influenced by the presence of debris and other cells in the sample. Typically, motility is calculated in three steps: sperm detection, tracking, and kinematic feature extraction, conducted as per World Health Organization (WHO) standards. However, these systems largely rely on traditional image processing techniques, which can be affected by noise, overlapping cells, and variations in image quality, leading to less reliable results. To overcome these limitations, this work proposes an AI-based framework. Rather than handling detection and tracking as independent processes, a joint object detection and tracking approach is used and trained AI models on low magnification bright field images, to improve sperm motility. The best trained model showed strong performance for sperm head detection, achieving an mAP of 0.896 and 0.492 at IoU 0.5 and 0.5–0.95, respectively. Sperm trackingperformance is evaluated with MOTA, notably achieved 0.595. This indicates that the framework is effective at tracking sperm heads, even in low-contrast microscopy images, and can be used for semen analysis. The framework is further validated with publicly available VISEM-tracking dataset achieving a pooled MOTA of 0.442. Moreover, to enhance prediction accuracy, a sliced image-based detection strategy is incorporated to test the identification of sperm cells even under complex visual conditions. The sliced inference slightly improved the recall, and missed detections and identity switches were also reduced, resulting in improved detection coverage and stable tracking. The models can be further optimized to improve efficiency; however, the results confirm that the methodology sounds in both modalities.
Zitieren
APA
Tariq, S., & Prasad, D. (2026). Joint Detection and Tracking of Sperm for Improving Motility Analysis. Nordic Machine Intelligence, 6(3), 30-34. https://doi.org/10.5617/nmi.13365
Vancouver
Tariq S, Prasad D. Joint Detection and Tracking of Sperm for Improving Motility Analysis. Nordic Machine Intelligence. 2026;6(3):30-34. doi:10.5617/nmi.13365
BibTeX
@article{Tariq2026,
author = {Tariq, Saher and Prasad, Dilip},
title = {Joint Detection and Tracking of Sperm for Improving Motility Analysis},
journal = {Nordic Machine Intelligence},
year = {2026},
volume = {6},
number = {3},
pages = {30--34},
doi = {10.5617/nmi.13365}
}
EUVIMED