Validation of an Artificial-Intelligence (AI)-Based Software for 3D Real-Time Ultrasound Tissue Response Monitoring in Liver Tumor Microwave Ablation-a single centre prospective study

Tuesday, April 1, 2025

3:18 PM – 3:27 PM CT

Location: Music City Center, 202 BC

Presenting Author(s)

Michal Mauda-Havakuk, MD, PhD

Director, Interventional Oncology services
Tel-Aviv Sourasky Medical Center, Israel

Author/Co-author(s)

RN

Rina Neeman, MD

Director MR unit TASMC
TASMC, Israel
YA

Yoav Amitai, MD

Attending abdominal and breast imaging
TASMC, Israel
OD

Orly Dvora, MA

Research assistant
TASMC, Israel

Purpose: Validation of a 3D real-time ultrasound (US) AI image analysis software (SW) for liver tumor microwave ablation (MWA) conformation, and correlation with contrast enhanced computed tomography (CECT) scans performed immediately (TO), 24-hours (T24), and 1 month (T1M) post ablation.

Materials and Methods: A total of 22 patients undergoing US monitored MWA with single antenna were included under clinical trial approved by our institutional committee. Ablations were performed under general anaesthesia. A tracker was placed on the sternum and CECT was acquired (pre-CECT). Microwaves antennas (Emprint, Medtronic) were positioned under CT or US guidance. US imaging (GE LOGIQ E10, probe C6-1) was tracked and registered to the pre-CECT. Ablation power was set at 100W and duration set according to tumor size. US imaging was tracked and registered to the pre-CECT imaging. During the ablation, continuous probe sweeping covering the entire ablation (AZ) zone was performed. US cine clips were recorded and analyzed by the SW to create a 3D AZ map. Two CECT scans with similar acquisition parameters were acquired at T0 and T24 and were compared to the SW AZ map using Dice Coefficient. T1M post ablation CECT was also acquired.

Results: Median age was 67 years (range 40-79). Tumors included 45%HCC tumors and the remaining various metastases. Median tumor diameter was 16 mm (range 9-30). AZ volume was segmented with average of 22.8, 32.1 and 22.1mL at T0, T24 and T1M respectively. Complete US intra-procedural data acquisition and analysis using the SW was completed for n=10. Dice coefficient between SW AZ map and T24 AZ was 79.1% (range 71.2%-85.9%), significantly higher (pv < 0.05) than between T0 and the T24 at 73.6% (range 65%-85%). For another n=12 cases, results were inadequate due to algorithmic challenges related to low image quality and variations in probe sweeping directions, necessitating further algorithm adjustments.

Conclusion: Use of the intra-procedural 3D US SW can accurately estimate the T24 CECT ablation zone. Our results suggest a novel tumor zone ablation conformation method, with real-time accuracy without radiation exposure.