Application of Deep Learning for Real-Time Ablation Zone Measurement in Ultrasound Imaging

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiofrequency Ablation

Background:

• Accurate delineation of ablation zones (AZs) is critical for evaluating radiofrequency ablation (RFA) therapy.
• Current manual measurement methods for AZs are prone to variability and inaccuracies.

Purpose of the Study:

• To assess the effectiveness of Artificial Intelligence (AI) in automating AZ measurements from ultrasound images.
• To compare the accuracy of AI-driven AZ measurements against manual measurements.

Main Methods:

• An in vitro study utilized chicken breast and liver samples undergoing bipolar RFA.
• Ultrasound images were captured periodically, and the Mask2Former AI model was employed for AZ segmentation.
• Measurements from AI and manual methods were compared, focusing on short-axis metrics.

Main Results:

• 308 RFA procedures yielded 7275 ultrasound images across liver and chicken breast tissues.
• AI and manual measurements for ablation zone diameters showed no significant differences (correlation coefficients > 0.96, p < 0.001).
• Bland-Altman plots and Deming regression indicated close alignment between AI and manual measurements (average differences of -0.259 mm for liver, -0.243 mm for chicken breast).

Conclusions:

• The Mask2Former AI model is validated as a promising tool for automated AZ measurement in RFA.
• This automation represents a significant advancement in reducing variability in RFA research measurements.