Study on microwave ablation temperature prediction model based on grayscale ultrasound texture and machine learning

Area of Science:

• Medical Physics
• Oncology
• Machine Learning

Background:

• Accurate temperature prediction is vital for effective microwave ablation in liver cancer treatment.
• Estimating the coagulation zone is essential for successful clinical outcomes.

Purpose of the Study:

• To develop a machine learning model for precise temperature prediction in microwave ablation.
• To assess the model's capability in determining the coagulation zone size.

Main Methods:

• Random forests machine learning model trained on ultrasound grayscale images and temperature data from porcine liver tissues.
• Experiments conducted at 15 W and 20 W ablation powers.
• Model accuracy validated by comparing predicted coagulation areas with actual measurements.

Main Results:

• The model achieved high accuracy in temperature prediction with average absolute errors of 1.14°C (15 W) and 4.73°C (20 W).
• Coagulation area measurement accuracy surpassed traditional ultrasound imaging, with error ratios of 0.402 and 0.182.
• The model identified key temperature-correlated texture features, offering interpretability.

Conclusions:

• The proposed temperature prediction model is suitable for real-time monitoring during microwave ablation.
• The model effectively assesses the coagulation zone range, improving treatment precision.