Imaging Informatics for Thermal Ablation: CT Thermometry-Based Tissue Injury Mapping

Sparse CT acquisition during CT-guided thermal ablation limits direct estimation of cumulative ther mal injury, which depends on temperature history rather than endpoint temperature alone. This study developed an imaging-informatics pipeline for near real-time Arrhenius injury mapping from workflow-compatible endpoint CT thermometry. Endpoint temperature fields were derived from ROI-level HU-temperature calibration and deployed in baseline-referenced form. Protocol-specific normalized temperature-history templates reconstructed heating trajectories at 180 and 300 s, and precompiled endpoint temperature-to-logΩ mappings avoided voxel-wise temporal integration at runtime. Validation was limited to ex vivo liver specimens, using porcine tissue for model de velopment and validation and two bovine cases for ex ploratory cross-tissue evaluation. Across five indepen dent experiments, per-update latency was 793.7-1688.0 ms (1198.5±434.3 ms). The logΩ ≥ 0 delineation yielded mean absolute major- and minor-axis errors of 0.96±0.73 mm and 1.98 ± 1.37 mm, respectively, with Dice 0.9151 ± 0.0311 and HD952.07±0.27 mmrelative to gross specimen boundaries. Dice and HD95 quantify post-alignment two-dimensional contour agreement after scale normalization rather than absolute localization accuracy. Leakage-controlled calibration analysis indicated limited out-of-sample CT thermometry performance, whereas two-level validation supported the stability of the endpoint-temperature-to-injury mapping. These ex vivo results demonstrate protocol-specific feasibility and motivate independent thermometry validation, uncertainty propagation, and prospective in vivo evaluation.