In minimally invasive liver tumor treatment without laparotomy, accurate registration between pre- and post-treatment images is essential for evaluating therapeutic effectiveness. Although contrast-enhanced ultrasound is increasingly used postoperatively due to its real-time capability, aligning it precisely with pre-treatment 3-D images remains challenging. Conventional methods often rely on magnetic sensors, which are costly and complex. Hence, a simpler and more affordable sensor-free method is needed. We propose a registration system that uses only ultrasound imaging, a common acoustic diagnostic tool. Before treatment, a robotically controlled probe scans the abdomen, and a 3-D vascular model is built from vessels visualized in ultrasound. Cross-sectional images in various orientations are generated from this model and used to train a deep-learning model. Post-treatment ultrasound images undergo vessel extraction, and the trained model estimates their acquisition positions. Based on this, initial alignment is performed with the 3-D model, followed by rigid point cloud registration to refine it. In evaluations using an abdominal phantom, 85.2% of 3000 test images were registered within a 10-mm error. Average processing time per image was 0.0782 s, demonstrating feasibility for real-time use. This method leverages the real-time nature of ultrasound, offering a promising sensor-free approach for convenient evaluation of therapeutic effectiveness.
KASAGI et al. (2025) studied this question.