HOPE-Net is a Mamba-Guided Hybrid Omni-Perceptual Enhanced Network for efficient and accurate sperm detection in clinical intracytoplasmic sperm injection (ICSI) procedures. Designed to overcome the limitations of manual assessment and existing deep learning approaches, HOPE-Net integrates state space models with convolutional operations to achieve global feature modeling with linear computational complexity, enabling real-time performance in complex clinical environments.
This repository contains the implementation of the method described in our paper:
HOPE-Net: Mamba-Guided Hybrid Omni-Perceptual Enhanced Network for Sperm Detection
-
Efficient Architecture: Only 1.9M parameters with inference speed of 48.6 FPS
-
Mamba-Guided Backbone: Mamba-Guided ShuffleNet (MGS-Net) with Parallel Mamba Aggregation Blocks (PMAB) for comprehensive feature fusion
-
Enhanced Neck Network: Hybrid State Space Path Aggregation Network (HSS-PAN) with Ghost Mamba Blocks (GMB) for cross-scale information interaction
-
Robust Performance: Evaluated on multiple datasets (HSDD, SDTB, SVIA, VISEM) with SOTA results
-
Clinical Applicability: Designed specifically for real-time sperm detection in challenging clinical scenarios
We are committed to promoting reproducible research. The complete source code and detailed documentation will be made publicly available upon formal acceptance of our paper.
🕐 Please check back later for updates, or watch this repository to be notified when the code is released.
For any questions regarding the paper or this implementation, please feel free to contact the authors.
📩 Email: 15563866837@163.com
🌟 We appreciate your interest in our work!