Development and Validation of an Algorithm Model for Predicting Heat Sink Effects during Pulmonary Thermal Ablation

Area of Science:

• Medical imaging and artificial intelligence
• Oncology and interventional radiology
• Computational modeling in medicine

Background:

• The heat sink effect significantly impacts thermal ablation efficacy for lung tumors.
• Current methods for predicting this effect are manual, time-consuming, and lack precision.
• An automated algorithmic approach is needed to improve accuracy and efficiency in surgical planning.

Purpose of the Study:

• To develop a convolutional neural network (CNN) model for automated segmentation of pulmonary nodules and vasculature.
• To accurately measure the distance between nodules and surrounding blood vessels.
• To create an algorithm model for predicting the intraoperative heat sink effect during thermal ablation.

Main Methods:

• Utilized Faster RCNN for nodule detection and VSPP-NET for segmentation of nodules and vasculature.
• Trained the algorithm on lung CT images from 392 patients, with validation and testing on separate cohorts.
• Compared the algorithm's heat sink effect predictions against expert manual segmentation.

Main Results:

• Achieved high recall (>0.88) and precision (>0.78) in pulmonary CT vasculature segmentation.
• Reduced average image segmentation time from 158 seconds (manual) to 29 seconds (automated).
• Demonstrated no significant difference in heat sink effect prediction between the algorithm and expert groups (p=0.687).

Conclusions:

• The developed algorithm model accurately predicts the heat sink effect in pulmonary thermal ablation.
• The model enhances speed and precision in nodule and vessel segmentation, saving planning time.
• Provides valuable data for surgeons to optimize ablation strategies and improve therapeutic results.