A dictionary learning approach for human sperm heads classification.

Area of Science:

• Biomedical Engineering
• Computer Vision
• Medical Diagnostics

Background:

• Semen analysis is crucial for diagnosing male infertility, with sperm morphology being a key factor.
• Manual assessment of sperm morphology is subjective and time-consuming, necessitating automated methods.
• Automatic sperm head classification presents challenges due to intra-class variations and inter-class similarities.

Purpose of the Study:

• To develop an automated method for classifying human sperm heads into four distinct classes.
• To utilize Dictionary Learning (DL) for constructing a robust dictionary of sperm head shapes.
• To improve the accuracy and objectivity of sperm morphology evaluation in semen analysis.

Main Methods:

• Extracted square patches from sperm head images for analysis.
• Employed Dictionary Learning (DL) to create class-specific dictionaries for sperm head shapes.
• Classified sperm heads by reconstructing image patches using learned dictionaries and evaluating reconstruction error.

Main Results:

• Achieved an average accuracy of 92.2% on the HuSHeM dataset.
• Attained an average recall of 62% on the SCIAN-MorphoSpermGS dataset.
• Demonstrated significant improvement over previous shape-feature-based classification methods.

Conclusions:

• Dictionary Learning (DL) is a highly effective method for classifying human sperm heads.
• The proposed DL approach offers a more balanced classifier with high precision and recall across all classes.
• Introduced a new dataset of human sperm head shapes to support future research in this area.