Effect of deep transfer and multi-task learning on sperm abnormality detection.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Reproductive Medicine

Background:

• Male infertility affects approximately half of all infertility cases worldwide.
• Current sperm morphology analysis by embryologists is subjective, time-consuming, and prone to bias.
• Objective and automated methods are needed to improve the accuracy and efficiency of sperm analysis.

Purpose of the Study:

• To develop and evaluate deep learning algorithms for automated sperm morphology analysis.
• To introduce a novel Deep Multi-task Transfer Learning (DMTL) approach for classifying sperm head, vacuole, and acrosome abnormalities.
• To establish a new benchmark in Sperm Morphology Analysis (SMA) using multi-task learning.

Main Methods:

• Proposed two deep learning algorithms: a network-based deep transfer learning approach and DMTL.
• DMTL combines deep transfer learning with multi-task learning for simultaneous classification of sperm parts.
• Algorithms were benchmarked using the freely-available MHSMA dataset.

Main Results:

• The proposed algorithms achieved state-of-the-art performance on accuracy, precision, and f0.5 metrics.
• Significant accuracy improvements were observed: 6.66% for head, 3.00% for acrosome, and 1.33% for vacuole.
• Achieved high accuracies of 84.00% (head), 80.66% (acrosome), and 94.00% (vacuole).

Conclusions:

• Deep learning, particularly DMTL, offers a robust and accurate solution for automated sperm morphology analysis.
• The developed algorithms can assist health institutions like fertility clinics in diagnosing male infertility more effectively.
• This work pioneers the application of multi-task learning in the field of Sperm Morphology Analysis.