SpermYOLO: A Coordinated YOLO-Based Detector for Accurate and Efficient Sperm and Impurity Detection in Microscopic Images

Shengqi Chen; Zilin Wang; Xingyu Pan; Wenting Yu; Pengchao Deng; Guohua Wu

cs.CVSep 13, 2026

SpermYOLO: A Coordinated YOLO-Based Detector for Accurate and Efficient Sperm and Impurity Detection in Microscopic Images

Authors: Shengqi Chen, Zilin Wang, Xingyu Pan, Wenting Yu, Pengchao Deng, Guohua Wu

Organizations: School of Electronic Engineering, Beijing University of Posts and Telecommunications, 10 Xitucheng Road, Haidian District, Beijing 100876, China · School of Information and Communication Engineering, Zhongyuan University of Technology, 41 Zhongyuan Middle Road, Zhengzhou, Henan Province 450007, China

Abstract

Accurate sperm detection is essential for computer-assisted semen analysis, yet it remains challenging in microscopic images due to dense distributions, visually similar artifacts, and sperm-like impurities. In this paper, we propose SpermYOLO, a coordinated and compact YOLOv11-derived framework for joint sperm and impurity detection in microscopic images. SpermYOLO introduces four architectural improvements: C3k2-IDB for channel-wise discriminative feature extraction, D2SEM for spatial--spectral semantic enhancement, MFM for adaptive multi-scale feature fusion, and the DESD Head for detail-enhanced shared prediction. Experiments on the SVIA semen microscopic imaging benchmark show that SpermYOLO achieves 97.2% sperm AP and 75.4% impurity AP, outperforming generic detectors, dedicated sperm detection models, and improved YOLO variants. Compared with the baseline model, SpermYOLO improves sperm AP, impurity AP, $\mathrm{mAP}_{50}$ , and $\mathrm{mAP}_{50:95}$ by 1.6, 10.0, 5.8, and 2.7 percentage points, respectively, while preserving a lightweight model scale. Cross-scene evaluation on the SDTB testicular-biopsy microscopy benchmark shows that SpermYOLO remains effective with extremely small sperm targets and complex tissue backgrounds, achieving the highest $\mathrm{mAP}_{50}$ and $\mathrm{mAP}_{50:95}$ of 74.8% and 31.2%, respectively. Ablation studies and qualitative analyses further support these improvements by demonstrating the contributions of the proposed modules and showing more focused feature response patterns than the baseline model. These findings suggest that SpermYOLO is an effective and efficient approach for sperm detection in challenging microscopic imaging scenarios.

Figures & tables

Figure 1: Sperm AP vs. Impurity AP on the SVIA dataset. SpermYOLO achieves the best trade-off between the two classes, outperforming generic detectors, dedicated sperm detection models, and YOLO variants.

Figure 2: Overall Architecture of SpermYOLO

Figure 3: Structure of the C3k2-IDB Module

Figure 5: Structure of the MFM Module

Figure 7: Training and validation curves of the baseline model and SpermYOLO on the SVIA dataset.

	C3k2-IDB	D2SEM	MFM	DESD Head	$\mathrm{AP}_{S}$	$\mathrm{AP}_{I}$	$\mathrm{mAP}_{50}$	$\mathrm{mAP}_{50:95}$	P	R	F1	Params	GFLOPs	Weights
YOLOv11n (Baseline)					95.6	65.4	80.5	44.5	85.5	73.8	78.7	2.6M	6.3	5.2MB
Single-factor Ablation	$\checkmark$				96.7	68.7	82.7	45.2	86.3	75.2	79.9	2.9M	6.1	5.8MB
		$\checkmark$			96.2	69.8	83.0	43.4	87.1	74.8	79.6	2.6M	6.3	5.3MB
			$\checkmark$		94.7	68.0	81.4	42.8	77.6	78.9	78.3	2.6M	6.3	5.2MB
				$\checkmark$	96.1	74.2	85.1	46.3	84.2	79.0	80.1	2.3M	6.0	5.0MB
Multi-factor Ablation	$\checkmark$			$\checkmark$	96.2	70.5	83.3	45.7	79.4	82.0	80.7	2.6M	5.8	5.6MB

Table 1: Ablation study of the proposed modules on the SVIA dataset.

Figure 8: Precision–recall curves of the baseline and SpermYOLO on the SVIA dataset.

Hidden Channels	$\mathrm{AP}_{S}$	$\mathrm{AP}_{I}$	$\mathrm{mAP}_{50}$	$\mathrm{mAP}_{50:95}$	P	R	F1	Params	GFLOPs	Weights
128	96.2	74.3	85.3	44.8	78.4	78.0	78.1	2.5M	4.9	5.2MB
256	97.2	75.4	86.3	47.2	84.2	81.6	82.7	2.6M	5.9	5.7MB
512	95.9	75.2	85.6	46.1	81.1	79.5	80.2	2.8M	9.8	7.3MB
1024	95.9	73.7	84.8	45.3	82.4	77.3	79.5	3.8M	25.0	13.6MB

Table 2: DESD Head hidden channel analysis on the SVIA dataset.

Configuration	$\mathrm{AP}_{S}$	$\mathrm{AP}_{I}$	$\mathrm{mAP}_{50}$	$\mathrm{mAP}_{50:95}$	P	R	F1	Params	GFLOPs	Weights
Backbone only	95.7	70.9	83.3	43.7	82.9	76.7	79.4	2.4M	6.0	5.2MB
Neck only	95.7	71.6	83.6	45.4	78.9	78.9	78.7	2.5M	6.0	5.5MB
Both	97.2	75.4	86.3	47.2	84.2	81.6	82.7	2.6M	5.9	5.7MB

Table 3: C3k2-IDB placement analysis on the SVIA dataset.

Category	Model	$\mathrm{AP}_{S}$	$\mathrm{AP}_{I}$	$\mathrm{mAP}_{50}$	$\mathrm{mAP}_{50:95}$	P	R	F1	Params	GFLOPs	FPS	Weights
Generic	Faster R-CNN [ 36 ]	93.2	49.4	71.3	37.5	71.2	68.3	69.4	41.1M	167.5	9.7	158.0MB
	SSD [ 29 ]	82.5	46.4	64.5	23.0	75.7	58.7	65.0	23.9M	30.5	43.2	91.1MB
	YOLOv5n	94.2	64.8	79.5	42.4	80.3	73.2	76.3	2.5M	7.1	98.0	5.0MB
	YOLOv6n [ 27 ]	91.6	62.1	76.9	34.3	79.8	74.0	76.5	4.2M	11.8	89.0	8.3MB
	YOLOv8n [ 43 ]	95.4	66.6	81.0	44.5	82.4	73.0	76.7	3.0M	8.1	96.0	6.0MB
	YOLOv10n [ 44 ]	95.3	67.9	81.6	41.4	76.9	78.2	77.4	2.3M	6.5	161.0	5.5MB

Table 4: Comparison with representative detectors on the SVIA dataset.

Category	Model	$\mathrm{mAP}_{50}$	$\mathrm{mAP}_{50:95}$	P	R	F1
Generic	Faster R-CNN [ 36 ]	14.4	4.7	17.2	43.8	24.7
	SSD [ 29 ]	12.4	3.5	26.9	15.7	19.8
	YOLOv5n	67.8	28.5	71.5	60.4	65.5
	YOLOv6n [ 27 ]	54.5	22.0	53.9	57.9	55.8
	YOLOv8n [ 43 ]	71.8	29.9	74.5	64.7	69.2
	YOLOv10n [ 44 ]	70.9	29.4	73.3	63.1	67.8

Table 5: Comparison with representative detectors on the SDTB dataset.

Figure 9: Qualitative detection comparison between the baseline and SpermYOLO on representative SVIA images. The upper row shows baseline predictions, and the lower row shows SpermYOLO predictions on the same images. Green, red, and blue boxes denote correct detections, false-positive detections, and missed targets, respectively.

Figure 10: Grad-CAM++ heatmap analysis of the baseline and SpermYOLO. From left to right, each row shows the original image, the baseline heatmap, and the SpermYOLO heatmap; warmer colors indicate stronger response.

Object Detection

Object detection locates and classifies objects in images or video, balancing localization accuracy, robustness, and inference speed.

See all 624 papers

Deep Learning for Semen Analysis in Male Infertility: Computer Vision, Multimodal Fusion, and Clinical Translation
Jul 6, 2026Runwei Guan, Shaofeng Liang, Jiacheng Weng +10Multimodal Fusion Microscopy Image Analysis
Thrust of AI, Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China · Department of Oncology, Suzhou Xiangcheng People’s Hospital, Suzhou, China · FertiTech AI, Shanghai, China +7
Interpretable Sperm Morphology Classification via Attention-Guided Deep Learning
Jun 18, 2026Zahra Asghari Varzaneh, Reza Khoshkangini, Thomas Ebner +1Explainable Medical Image Analysis Microscopy Image Analysis
Department of Computer Science and Media Technology, Malmö University, Malmö, Sweden · Kepler Universitätsklinikum, Linz, Austria · NewLifeAid-Global AB, Sweden
Center-Aware Detection with Swin-based Co-DETR Framework for Cervical Cytology
Apr 2, 2026Yan Kong, Yuan Yin, Hongan Chen +2Object Detection Medical Image Localization
School of Intelligence Science and Technology, Nanjing University, Nanjing, China. · School of Biomedical Engineering, ShanghaiTech University, Shanghai, China. · State Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, China.

Jul 6, 2026J/K move · Enter open · S save

Deep Learning for Semen Analysis in Male Infertility: Computer Vision, Multimodal Fusion, and Clinical Translation

Runwei Guan, Shaofeng Liang, Jiacheng Weng, Xiaoyi Gu+9

Thrust of AI, Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China · Department of Oncology, Suzhou Xiangcheng People’s Hospital, Suzhou, China · FertiTech AI, Shanghai, China+7

Multimodal Fusion Microscopy Image Analysis

Open paper AskPDF arXiv

Male infertility contributes substantially to the global infertility burden, and sperm analysis remains central to diagnosis, treatment planning, and assisted reproductive technology. Conventional semen evaluation, however, is labor-intensive, operator-dependent, and limited by inter- and intra-observer variability, motivating the development of objective and reproducible computational approaches. This review provides a comprehensive and perspective-oriented synthesis of artificial intelligence-driven sperm analysis, with a focus on computer vision, deep learning, multimodal fusion, robustness, and clinical translation. We first review task-specific methods for sperm detection and counting, tracking-based motility assessment, semantic and instance segmentation, morphology and defect classification, functional assessment, and genetic integrity evaluation. We then summarize public datasets, benchmarks, evaluation metrics, and emerging multimodal strategies that integrate microscopic images, time-lapse videos, CASA-derived parameters, DNA integrity assays, and clinical metadata. Beyond algorithmic performance, we discuss key barriers to real-world deployment, including data scarcity, cross-center domain shift, annotation inconsistency, interpretability, uncertainty calibration, privacy-preserving learning, and workflow integration. Finally, we outline a staged clinical translation roadmap spanning technical standardization, multicenter retrospective validation, silent prospective evaluation, human-in-the-loop clinical testing, ART outcome validation, regulatory approval, and post-market monitoring. By organizing the field from task-specific visual recognition to trustworthy multimodal reproductive intelligence, this review highlights both the progress and the unresolved challenges required to translate AI-driven sperm analysis into clinically meaningful decision support.

Abstract

Object Detection

Explore similar work

Deep Learning for Semen Analysis in Male Infertility: Computer Vision, Multimodal Fusion, and Clinical Translation

Interpretable Sperm Morphology Classification via Attention-Guided Deep Learning

Center-Aware Detection with Swin-based Co-DETR Framework for Cervical Cytology