A novel deep learning method for automatic assessment of human sperm images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Reproductive Medicine

Background:

• Sperm morphology analysis (SMA) is crucial for diagnosing male infertility.
• Existing methods may not adequately analyze all sperm components, such as the acrosome.
• Accurate SMA is vital for assessing male reproductive health.

Purpose of the Study:

• To develop a novel deep learning algorithm for detecting morphological abnormalities in human sperm cells.
• To create a standardized benchmark dataset (MHSMA) for machine learning in sperm morphology analysis.
• To evaluate the algorithm's performance on head, acrosome, and vacuole abnormalities, including its ability to handle non-stained and low-resolution images.

Main Methods:

• Development of a deep neural network architecture for sperm image analysis.
• Creation and utilization of the MHSMA dataset (1,540 images from 235 patients).
• Implementation of data augmentation and sampling methods to address data imbalance.

Main Results:

• The algorithm achieved high F0.5 scores: 84.74% for acrosome, 83.86% for head, and 94.65% for vacuole abnormalities.
• Outperformed state-of-the-art methods in acrosome and vacuole abnormality detection on the MHSMA benchmark.
• Demonstrated real-time classification capabilities on standard hardware, enabling rapid analysis.

Conclusions:

• The proposed deep learning algorithm offers a highly accurate and efficient solution for sperm morphology analysis.
• The MHSMA dataset provides a valuable public resource for advancing machine learning in male infertility research.
• This technology has the potential to significantly aid embryologists in clinical decision-making for fertility treatments.