Advanced Multi-Level Ensemble Learning Approaches for Comprehensive Sperm Morphology Assessment

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Healthcare
• Computational Biology

Background:

• Sperm morphology is crucial for male infertility diagnosis and assisted reproductive technologies like IVF and ICSI.
• Manual evaluation of sperm morphology is subjective and inconsistent, necessitating automated solutions.
• Accurate sperm morphology assessment aids in reproductive healthcare decision-making.

Purpose of the Study:

• To develop a robust, fully automated framework for classifying sperm morphology.
• To minimize observer variability in sperm morphology assessment.
• To improve diagnostic support in male fertility evaluation.

Main Methods:

• An ensemble-based classification approach combining Convolutional Neural Network (CNN) features.
• Feature-level and decision-level fusion techniques using EfficientNetV2 variants.
• Classification via Support Vector Machines (SVM), Random Forest (RF), and Multi-Layer Perceptron with Attention (MLP-Attention), with soft voting for decision fusion.

Main Results:

• The ensemble framework achieved 67.70% accuracy on the Hi-LabSpermMorpho dataset (18 classes).
• The fusion-based model significantly outperformed individual classifiers.
• Ensemble techniques and multi-CNN integration addressed class imbalance and improved generalizability.

Conclusions:

• The proposed methodology offers a substantial improvement over traditional and single-model approaches for automated sperm morphology classification.
• Ensemble learning and multi-level fusion provide a reliable and scalable solution for clinical male fertility assessment.
• This automated system enhances diagnostic accuracy and consistency in reproductive healthcare.