Scholarlink

<p><b>Background</b> Spinal surgical robots have achieved widespread clinical adoption over two decades, yet their safe and precise bone resection remains fundamentally limited by the lack of reliable real-time tissue discrimination capabilities. Conventional single-modality sensing approaches (imaging, force, acoustic, bioelectrical impedance) each suffer from inherent drawbacks that compromise surgical safety. This state-of-the-art review systematically evaluates AI-driven tissue discrimination sensing methods specifically developed for robot-assisted spine surgery.</p><p><b>Methods</b> A comprehensive literature search was conducted across PubMed, IEEE Xplore, and CNKI (1990–2025) following PRISMA guidelines. Sixty eligible studies were included and categorized by sensing modality. We performed a comparative analysis focusing on the interplay between surgical procedures, sensor technologies, and AI algorithm architectures, with particular emphasis on the clinical value of AI-enabled multimodal fusion.</p><p><b>Results</b> Three key insights emerged. First, surgical operations exhibit distinct AI-sensing pairing preferences: drilling uses force/acoustic signals with SVM/BP networks for real-time breakthrough detection; milling relies on force signals with LSTMs for continuous time-series processing; cutting employs imaging with CNNs for anatomical segmentation. Second, sensor type fundamentally dictates AI design—imaging sensors pair with CNNs for spatial features, while force/acoustic sensors use recurrent networks for temporal patterns. Third, AI-driven multimodal fusion’s core value is resolving single-modality blind spots, not incremental accuracy gains. Fusion of force and acoustic signals elevated transitional zone recognition from 62% to 92%, significantly improving clinical robustness.</p><p><b>Conclusion</b> Single-modality sensing has fundamental limitations, making AI-enabled multimodal fusion an inevitable development trend. We propose a decision tree framework for selecting optimal AI-sensing combinations based on surgical requirements. Future directions include integrated AI-sensor platforms, pathology-adaptive learning models, and standardized evaluation protocols to advance autonomous spinal surgery.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

AI-driven tissue discrimination sensing methods for robot-assisted spine surgery: a state-of-the-art review

  • 期刊论文

  • 发表日期：

    2026年4月25日

查看全文
  • Yaowen Nuo,
  • Xiuyuan Chen,
  • Hongxing Shen

摘要

Background Spinal surgical robots have achieved widespread clinical adoption over two decades, yet their safe and precise bone resection remains fundamentally limited by the lack of reliable real-time tissue discrimination capabilities. Conventional single-modality sensing approaches (imaging, force, acoustic, bioelectrical impedance) each suffer from inherent drawbacks that compromise surgical safety. This state-of-the-art review systematically evaluates AI-driven tissue discrimination sensing methods specifically developed for robot-assisted spine surgery.

Methods A comprehensive literature search was conducted across PubMed, IEEE Xplore, and CNKI (1990–2025) following PRISMA guidelines. Sixty eligible studies were included and categorized by sensing modality. We performed a comparative analysis focusing on the interplay between surgical procedures, sensor technologies, and AI algorithm architectures, with particular emphasis on the clinical value of AI-enabled multimodal fusion.

Results Three key insights emerged. First, surgical operations exhibit distinct AI-sensing pairing preferences: drilling uses force/acoustic signals with SVM/BP networks for real-time breakthrough detection; milling relies on force signals with LSTMs for continuous time-series processing; cutting employs imaging with CNNs for anatomical segmentation. Second, sensor type fundamentally dictates AI design—imaging sensors pair with CNNs for spatial features, while force/acoustic sensors use recurrent networks for temporal patterns. Third, AI-driven multimodal fusion’s core value is resolving single-modality blind spots, not incremental accuracy gains. Fusion of force and acoustic signals elevated transitional zone recognition from 62% to 92%, significantly improving clinical robustness.

Conclusion Single-modality sensing has fundamental limitations, making AI-enabled multimodal fusion an inevitable development trend. We propose a decision tree framework for selecting optimal AI-sensing combinations based on surgical requirements. Future directions include integrated AI-sensor platforms, pathology-adaptive learning models, and standardized evaluation protocols to advance autonomous spinal surgery.