Related Experiment Video
Updated: Oct 21, 2025
08:48
Flow Cytometric Analysis of Biomarkers for Detecting Human Sperm Functional Defects
Published on: April 21, 2022
2.8K
Multi-model CNN fusion for sperm morphology analysis
Mecit Yüzkat1, Hamza Osman Ilhan2, Nizamettin Aydin2
1Yildiz Technical University, Faculty of Electrical and Electronic, Department of Computer Engineering, Turkey; Mus Alparslan University, Faculty of Engineering and Architecture, Department of Computer Engineering, Turkey.
Computers in Biology and Medicine
|September 7, 2021
Summary
Automated sperm morphology analysis using deep learning improves accuracy in diagnosing male infertility. Convolutional neural networks (CNNs) offer objective, efficient semen analysis, crucial for identifying causes of infertility.
Area of Science:
- Reproductive Medicine
- Artificial Intelligence in Healthcare
- Medical Imaging Analysis
Background:
- Male infertility affects 40% of infertile couples, with semen analysis being the primary diagnostic tool.
- Manual semen analysis is time-consuming and prone to observer variability, necessitating objective, automated methods.
- Deep learning, particularly Convolutional Neural Networks (CNNs), excels at analyzing large image datasets for complex pattern recognition.
Purpose of the Study:
- To develop and evaluate automated sperm morphology classification systems using deep learning.
- To compare the performance of six distinct CNN models and decision-level fusion techniques for sperm image analysis.
- To achieve objective and highly accurate classification of sperm morphology from diverse datasets.
Main Methods:
- Six Convolutional Neural Network (CNN) models were designed for automated sperm image classification.
- Decision-level fusion techniques (hard-voting and soft-voting) were applied to combine the outputs of the CNN models.
- Three public sperm morphology datasets (SMIDS, HuSHeM, SCIAN-Morpho) were utilized, with cross-validation for robust evaluation.
- Data augmentation and mini-batch analysis were employed to enhance classification performance.
Main Results:
- The soft-voting fusion approach achieved high classification accuracies: 90.73% on SMIDS, 85.18% on HuSHeM, and 71.91% on SCIAN-Morpho.
- The proposed deep learning models demonstrated successful automated classification across multiple datasets.
- The system provided objective and efficient analysis, overcoming limitations of manual semen evaluation.
Conclusions:
- Automated sperm morphology classification using CNNs and fusion techniques is a viable and effective approach for diagnosing male infertility.
- The developed system offers a significant improvement in objectivity and efficiency over traditional manual semen analysis.
- This AI-driven method holds promise for enhancing the accuracy and reliability of male infertility assessments in clinical settings.
