Multi-source information fusion using CNN-LSTM-Attention for bone layer recognition in robotic orthopedic grinding

Area of Science:

• Robotics in Surgery
• Artificial Intelligence in Orthopedics
• Biomedical Signal Processing

Background:

• Epiphyseal opening for bony bridge resection demands high precision, which is currently limited by traditional surgical methods.
• Existing robot-assisted surgeries lack intraoperative decision-making for accurate bone localization and grinding.
• Distinguishing between bone types (cortical, cancellous, idling) in real-time is crucial for safe and effective robotic orthopedic procedures.

Purpose of the Study:

• To develop a multi-source information fusion framework for real-time bone layer differentiation during robotic orthopedic grinding.
• To enhance the precision and autonomy of robot-assisted surgeries by enabling real-time identification of bone states.
• To improve the safety and efficacy of epiphyseal opening procedures through accurate localization and grinding.

Main Methods:

• A CNN-LSTM-Attention network was developed for real-time differentiation of bone layers (idling, cancellous, cortical) using acceleration, force, and acoustic signals.
• The network analyzes the mapping between sensor signals and bone density to recognize different bone states during robotic grinding.
• A dataset was created from independent grinding trials, employing trial-wise splitting to prevent data leakage and ensure robust model evaluation.

Main Results:

• The proposed CNN-LSTM-Attention framework achieved a test accuracy of 95.06% ± 0.53%, significantly outperforming baseline models.
• The model demonstrated real-time feasibility with inference latencies of 2.8 ms (CPU) and 1.2 ms (Jetson Orin), well within robot control cycle limits.
• Tri-modal signal fusion (acceleration, force, acoustic) yielded the best performance, highlighting the benefit of multi-source information integration.

Conclusions:

• The developed multi-source information fusion framework enables accurate, real-time bone layer differentiation in robotic orthopedic grinding.
• This approach significantly enhances intraoperative localization and grinding precision, addressing limitations of current surgical techniques.
• The findings support the potential of AI-driven robotic systems to improve safety and outcomes in orthopedic surgeries requiring precise bone manipulation.