Classifier performance overview.

Background and objective:
Male fertility assessment through sperm morphology analysis remains a critical component of reproductive health evaluation, as abnormal sperm morphology is strongly correlated with reduced fertility rates and poor assisted reproductive technology outcomes. Traditional manual analysis performed by embryologists is time-intensive, subjective, and prone to significant inter-observer variability, with studies reporting up to 40% disagreement between expert evaluators. This research presents a novel deep learning framework combining Convolutional Block Attention Module (CBAM) with ResNet50 architecture and advanced deep feature engineering (DFE) techniques for automated, objective sperm morphology classification.
Materials and methods:
We propose a hybrid architecture integrating ResNet50 backbone with CBAM attention mechanisms, enhanced by a comprehensive deep feature engineering pipeline. The framework incorporates multiple feature extraction layers (CBAM, GAP, GMP, pre-final) combined with 10 distinct feature selection methods including Principal Component Analysis (PCA), Chi-square test, Random Forest importance, variance thresholding, and their intersections. Classification is performed using Support Vector Machines with RBF/Linear kernels and k-Nearest Neighbors algorithms. The model was rigorously evaluated on two benchmark datasets: SMIDS (3000 images, 3-class) and HuSHeM (216 images, 4-class) using 5-fold cross-validation.
Results:
The proposed framework achieved exceptional performance with test accuracies of 96.08 ± 1.2% on SMIDS dataset and 96.77 ± 0.8% on HuSHeM dataset using deep feature engineering, representing significant improvements of 8.08% and 10.41% respectively over baseline CNN performance. McNemar’s test confirmed statistical significance (). The best configuration (GAP + PCA + SVM RBF) demonstrated superior performance compared to existing state-of-the-art approaches, including recent Vision Transformer and ensemble methods.
Conclusions and clinical impact:
This research demonstrates the effectiveness of attention-based deep learning combined with sophisticated feature engineering for sperm morphology analysis. The proposed framework achieves state-of-the-art performance while providing clinically interpretable results through Grad-CAM attention visualization. Clinical implications include: (1) standardized, objective fertility assessment reducing diagnostic variability, (2) significant time savings for embryologists (from 30–45 minutes to <1 minute per sample), (3) improved reproducibility across laboratories, and (4) potential for real-time analysis during assisted reproductive procedures, ultimately enhancing patient care and treatment outcomes in reproductive medicine.