A fully automated hybrid human sperm detection and classification system based on mobile-net and the performance...

Area of Science:

• Reproductive Medicine
• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Sperm morphology is a key indicator of male fertility and a critical component of semen analysis.
• Manual assessment of sperm morphology is subjective and prone to human error, necessitating objective and automated methods.

Purpose of the Study:

• To develop and validate a fully automated smartphone-based hybrid system for sperm morphology analysis.
• To eliminate human subjectivity in semen analysis by automating sperm shape segmentation and classification.

Main Methods:

• A hybrid system combining automatic segmentation (clustering with group sparsity) and classification (machine learning and deep neural networks).
• Development of a novel, publicly available sperm image dataset for ground truth.
• Application of conventional features (wavelet transform, descriptors) and deep learning (Mobile-Net) for classification.

Main Results:

• Conventional features with Support Vector Machines achieved up to 83.8% accuracy.
• The Mobile-Net deep learning architecture achieved 87% accuracy in sperm classification.
• The hybrid system demonstrated effectiveness for mobile sperm morphology analysis.

Conclusions:

• A fully automated hybrid system integrating group sparsity for segmentation and Mobile-Net for feature extraction offers an effective mobile solution for sperm morphology analysis.
• The proposed system enhances objectivity and efficiency in fertility diagnostics.