Accurate localization of motor function using transcranial magnetic stimulation with segmentation-free head modeling in patients with brain tumors

<i>Objective.</i>Transcranial magnetic stimulation (TMS) is widely employed to diagnose neurological conditions and for preoperative functional brain mapping. However, the electric field (EF) distribution in the brain is often distorted by anatomical and electrical complexities, including brain tumors, which compromise localization accuracy. Conventional head models based on segmentation assume uniform tissue conductivity, which fails to capture tumor heterogeneity and introduces uncertainty into EF modeling. This study aimed to improve the accuracy of TMS-induced EF localization in motor area mapping by utilizing individualized segmentation-free head models. These models derive voxel-level conductivity estimates directly from magnetic resonance imaging (MRI) data, eliminating the need for tissue segmentation and accommodating tumor-specific variability.<i>Methods.</i>Four patients with intra-axial brain tumors near the motor eloquent areas underwent preoperative TMS mapping and intraoperative direct electrical stimulation (DES). Individualized head models were developed using segmentation-based and segmentation-free approaches. EF distributions and localization accuracy were evaluated using DES as the gold standard. The accuracy was quantified by calculating the Euclidean distances between the DES stimulation sites, TMS-derived hotspots, and center of gravity (CoG) of the EF distributions.<i>Results.</i>The segmentation-free model demonstrated superior localization accuracy with fewer stimulation samples, achieving a CoG-DES distance of 3.23 ± 0.45 mm. This CoG distance is less than half of the 7.78 ± 1.99 mm obtained with a navigated TMS. It effectively accounts for tumor heterogeneity by incorporating voxel-level conductivity variations, leading to more precise EF estimations than segmentation-based models. Individual variability was observed in the sample size required to achieve convergence of the estimated area, which was influenced by factors such as tumor location, size, progression, and electrical inhomogeneities. The segmentation-free model consistently required fewer samples across all subjects, demonstrating the efficiency of the computational processing.<i>Significance.</i>Segmentation-free head models significantly improved TMS localization accuracy with fewer stimulation samples, particularly in cases with tumor-induced cortical distortions. This approach is promising for broader clinical applications and warrants validation in larger and more diverse cohorts.