SHMC-Net: A Mask-guided Feature Fusion Network for Sperm Head Morphology Classification

• DOI: 10.48550/arxiv.2402.03697
• Publication Date: 2024-02-05
• Authors: Nishchal Sapkota; Yejia Zhang; Sirui Li; Peixian Liang; Zhuo Zhao; Danny Z. Chen
• Keywords: Feature (linguistics); Morphology; Net (polyhedron)

Male infertility accounts for about one-third of global cases. Manual assessment sperm abnormalities through head morphology analysis encounters issues observer variability and diagnostic discrepancies among experts. Its alternative, Computer-Assisted Semen Analysis (CASA), suffers from low-quality images, small datasets, noisy class labels. We propose a new approach classification, called SHMC-Net, which uses segmentation masks heads to guide the classification images. SHMC-Net generates reliable using image priors, refines object boundaries with an efficient graph-based method, trains network crops mask corresponding masks. In intermediate stages networks, features are fused fusion scheme better learn morphological features. To handle labels regularize training on applies Soft Mixup combine mixup augmentation loss function. achieve state-of-the-art results SCIAN HuSHeM outperforming methods that use additional pre-training or costly ensembling techniques.