Tsukuba's CNN cuts incision tracking error 2.5x in electrosurgery navigation

A team from the University of Tsukuba led by Yu Chun Wang and Yoshihiro Kuroda has developed a CNN-based method to trace incision trajectories during electrosurgery, a core need for image-guided navigation. Their approach uses thermal imaging to distinguish when an electric knife or ultrasonic cutter is actually contacting tissue, then extracts those contacting frames to build precise incision paths. In tests with electric knives, the CNN hit 97.2% classification accuracy, and with ultrasonic cutters it reached 93.7%.

The key advance is error reduction: the new method improves incision trajectory prediction by more than 2.5x over the conventional technique, while also avoiding the missing-trajectory failures common with convolutional long short-term memory (convLSTM) networks. By connecting the thermal intensity centroid of frames flagged as contacting, the system dynamically updates preoperative 3D models during surgery. This could give surgeons real-time feedback on where cuts have been made, improving accuracy and safety when navigating around subsurface structures. The paper, which appeared on arXiv on August 14, 2026, is currently in preprint and opens up a practical path for thermal-vision-based surgical navigation with minimal hardware overhead.