The abstract is read at the publisher; the summary is JClub's.
DeepoLITT predicts LITT temperature fields and CEM43 thermal dose from MRI tumor images with 0.04±0.05 s per 3D sample, versus 24.32±2.12 s by FEM.
- Why it matters: Patient-specific thermal dose prediction could improve MRI-guided laser therapy planning, but existing numerical solvers are computationally expensive and may be limited in accuracy and reliability.
- What they did: A branch-trunk deep operator network was trained on 400 FEM-generated synthetic samples (90% train/validation, 10% test) spanning noise levels 0–20% and two probe types (FullFire, SideFire). Spatial agreement on CEM43 isodose contours at 2, 10, and 60 minutes was quantified using Dice Similarity Coefficient and Hausdorff Distance.
- The result: For FullFire at noise-free conditions, DSCs were 0.96±0.02, 0.97±0.21, and 0.96±0.21 with HDs 0.09±0.08 mm, 0.07±0.07 mm, and 0.06±0.08 mm; SideFire showed DSCs 0.96±0.02, 0.97±0.21, and 0.96±0.21 with HDs 0.20±0.31 mm, 0.11±0.30 mm, and 0.15±0.31 mm, with DSC decreasing and HD increasing as noise rose.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.