Fiber-optic Temperature Field Monitoring Based on Deep Learning Network for Tumor Thermal Ablation

Thermal ablation therapy plays a pivotal role in cancer treatment. Nevertheless, the lack of precise monitoring techniques for tracking the ablation site's position and temperature distribution poses challenges. This article introduces a fiber-optic approach based on deep learning to real-time invert the three-dimensional (3D) position and temperature of the ablation site, and the effective temperature impact zone. It involves designing a flexible fiber-optic sensing film with conformal ability to map the temperature field on the tissue's surface. An inverse algorithm employing the Gated Recurrent Unit model accurately estimates the three-dimensional location and temperature distribution. Experiments validate the Gated Recurrent Unit model and compare its accuracy with other deep learning networks. Results demonstrate the proposed method's effectiveness, with accuracy below 0.4 mm for position estimation, below 0.11 mm for influence zone radius, and below 0.5 ℃ for temperature estimation. The algorithm exhibits high computational efficiency with a test time of 124 ms per dataset. This fiber-optic sensing method, combined with deep learning, shows promise for online monitoring of the internal three-dimensional temperature field.