Purpose: Minimally invasive techniques are used as an alternative option to conventional surgery for liver cancer patients, but local recurrence has been relatively high due to insufficient local tumor ablation. The purpose of our work is to address the tumor-targeting limitations of 2D ultrasound (US) by developing a novel intraoperative 3DUS system for improved image guidance and verification of therapy applicators during liver tumor ablation with minimal cost and changes to current workflow. We have developed and previously reported on a three-motor mechanical 3DUS scanner. Here we report on its integration and evaluation of the complete 3DUS-based mechanical guidance system and a deep learning 2D therapy applicator segmentation algorithm for image guidance during focal liver tumor ablation therapy.
Methods: Our mechanical scanner for variable 3D US fields-of-views is held by a portable counterbalanced mechanical system with 5 encoders to track the position of a 2D US transducer. Optical tracking of a mounted stylus was performed to evaluate the error in the pose determined by the mechanically encoded system. A U-Net was developed and trained on 187 needle-like images from insertions in phantoms, liver, gynecological, prostate, and kidney procedures with 40 liver ablation applicator images used for evaluation.
Results: Optical tracking of the encoded system resulted in a mean error of 1.9Â±1.3 mm when combining the 5 motions of the tracked joints. Applicator segmentation resulted in 2.9Â±4.0 mm mean tip error and 0.9Â±0.7 Â° mean trajectory error for insertions greater than 25 mm.
Conclusion: The proposed mechanically assisted 3D US guidance system is able to track the position of a 2D US transducer accurately and a deep learning algorithm is able to identify applicators in 2D US images to increase image guidance during focal liver tumor therapy. Current work is focused on performing a simulated image-guided phantom procedure.
Funding Support, Disclosures, and Conflict of Interest: The authors are grateful for the funding support from the Ontario Institute of Cancer Research (OICR), the Canadian Institutes of Health Research (CIHR), and by the Natural Sciences and Engineering Research Council of Canada (NSERC).