Key points are not available for this paper at this time.
Abstract PURPOSE To plan laser probe trajectories for magnetic resonance imaging (MRI)-guided laser induced thermal therapy (MRIgLITT), current clinical practice relies on manual inspection on MRI images to find a path with minimum safety risk and maximal treatment impact. This process is empirical and time-consuming with potentially suboptimal solutions. The study aims to develop an automated trajectory optimization algorithm for reducing planning time and improving patient safety. METHODS Seven brain tumor cases treated with MRIgLITT were retrospectively reviewed. For each case, preoperative MRIs were auto-segmented for 16 brain tissue types (e.g., skull, artery, vein, etc.) using a deep learning-based head modeling algorithm (Sim4life v8.0, ZMT Zurich MedTech AG, Switzerland); Then an optimize trajectory was generated by searching potential trajectories based on a priori established criteria, such as trajectory feasibility (angle to the skull, not entering in the face or neck, etc.), safety risk (avoiding blood vessels and optical cranial nerves, etc.), and expected treatment efficacy (100% tumor coverage with assumed 1.5 cm laser radiation radius). The predicted optimal trajectories were finally compared with the trajectories prescribed from clinical procedures, and differences were analyzed descriptively. RESULTS For each case, the algorithm took approximately 4 minutes for brain tissue segmentation and 2 minutes for trajectory optimization on a desktop computer. The predicted trajectories from 7 cases were visually matched with prescribed trajectories with offsets of entry point (9.94±5.97 mm), target point (4.59±2.23 mm) and trajectory length (5.90±3.69 mm). Qualitative review of these trajectories confirmed their adherence to the above principles, meaning they would likely have been usable for their respective cases. CONCLUSION We demonstrated the feasibility and efficiency of an automated trajectory optimization algorithm for MRIgLITT. Future studies will improve the performance of brain tissue segmentation, assess anticipated versus achieved tissue treatment volumes, and prospectively evaluate application into case planning.
Lu et al. (Thu,) studied this question.