Parkinson’s disease (PD) is a progressive neurodegenerative disorder that primarily affects motor function, significantly impairing patients’ quality of life. Current treatment options, while effective in managing symptoms, often become less efficient over time, prompting the need for innovative therapeutic approaches (Hwu et al., 2021). Convection-enhanced drug delivery (CED) has emerged as a promising strategy for targeted drug administration in PD, particularly for cell and gene therapy. CED infuses therapeutic agents via stereotactically placed catheters under positive pressure to drive bulk flow through the interstitium, achieving high local concentrations over larger brain volumes, reducing systemic side effects and improving local drug concentration (Lam et al., 2011; Barua et al., 2014). In practice, distribution is constrained by anatomical variability, tissue anisotropy, and potential reflux along the catheter track, which can challenge precise targeting (Lam et al., 2011; Barua et al., 2013; Larson, 2021).

The putamen, as part of the basal ganglia, plays a key role in motor control and is a central target for CED in PD therapy (Miranpuri et al., 2012). Conventional CED methods, however, often target the entire putamen, resulting in inefficient drug use and unnecessary exposure of non-motor regions. This broad targeting not only increases treatment costs by requiring more of the drug but also side effects by affecting regions that do not contribute to motor function. Thus, more precise drug delivery methods that target specific, motor-relevant areas of the putamen are needed to improve therapeutic efficacy and reduce unnecessary drug use. Prior work using similar structural or multimodal approaches, such as combining resting-state fMRI and diffusion MRI tractography, have delineated putaminal subregions with distinct connectivity (Draganski et al., 2008; Tian et al., 2020). Analogous to the principle of applying deep brain stimulation (DBS) for PD in the dorsolateral motor part of the subthalamic nucleus (STN) to achieve the optimal motor improvement, these findings suggest an opportunity to focus CED therapy on motor-connected segments rather than the entire putamen.

The goal of this study is to use existing connectivity knowledge to explore simulation prior to CED and inform the planning. In a clinical workflow, an easily identifiable predefined parcellation of the putamen is applied, then diffusion-weighted imaging (DWI) tractography maps structural connectivity onto these segments to verify the expected motor bias in the putamen. On this basis, the feasibility of connectivity guided CED is evaluated by using two planned trajectories and an in-house developed stepwise-injection simulation that quantifies target coverage. A genetic algorithm optimizes a theoretical stepwise infusion protocol to increase coverage of the selected segment while respecting safety constraints. The distribution model is kept simple to keep assumptions explicit and computations tractable. Simulation outputs are therefore interpreted as planning aids—to quantify safety, coverage and to support case-specific decisions—rather than predictors of in vivo distribution. While clinical validation is not yet available for the present application, MRI-based presurgical planning tools in other neurosurgical settings have been evaluated prospectively and shown to improve perioperative outcomes and functional status compared with conventional planning (Yang et al., 2009). The objective is to assess the clinical feasibility of connectivity-guided CED in the putamen with safety and coverage metrics across the two trajectories.

A total of 20 patients were included in the analysis. Fourteen patients (70%) were male and six (30%) were female. The mean age at the time of implantation was 59.3 ± 8.9 years (range 40.5–70.1 years). According to the Hoehn & Yahr scale, 11 patients (55.0%) were classified as stage III, 4 (20.0%) as stage II, 4 (20.0%) as stage IV, 1 (5.0%) as stage I.

We aimed to characterize how motor- and non-motor-associated structural connectivity distributes across a simple, AC-based segmentation of the putamen that is compatible with clinical workflows, and to examine whether this organization can be used to support presurgical planning for convection-enhanced drug delivery (CED) into a motor-dominated segment of the putamen.

In our cohort, motor-associated cortical and subcortical regions showed higher connectivity strength to the superior posterior putamen segment, while non-motor regions favored other segments. The motor-associated regions such as 3a/b, 4a/p, 6mp SMA, 6ma pre-SMA, the cerebellum, and the STN displayed a higher connectivity strength to the superior posterior segment of the putamen. Particularly, the 6mp SMA showed almost exclusive structural connectivity to this segment. Other regions, such as 3a/b, were more generally connected to both posterior segments, while the cerebellum and the STN were primarily linked to both superior segments of the putamen. Non-motor-associated regions exhibited different patterns. In particular, the insula region connects predominantly to the superior anterior segment and the amygdala almost entirely to the posterior inferior segment. Based on this pattern, we selected the superior posterior segment as the CED target. Using stereotactic planning and a simple infusion simulation, both frontal and occipital drug injection trajectories could be planned safely in most hemispheres (85% vs. 82.5% safe; mid-p McNemar p = 0.8462) and achieved comparable target-segment coverage, with no statistically significant differences between them (7.69% vs. 8.74%; paired t-test p = 0.4875). The findings were obtained using deterministic tractography implemented in CE-marked clinical software and indicate that even a coarse, AC-based segmentation can distinguish a superior posterior segment with predominantly motor-associated connectivity from other segments with more limbic or paralimbic connectivity.

The tractography and simulations presented here are intended as a planning adjunct. They provide a reproducible way to select a motor-dominant putaminal segment using a simple stereotactic segmentation and compare candidate trajectories under explicit safety constraints using a transparent geometric coverage metric. A prospective validation would be required for claiming any clinical benefits. At the same time, MRI-based presurgical simulation tools in other neurosurgical settings have been evaluated prospectively with measurable clinical benefit. For example, in a randomized controlled study of MRI-based VR presurgical planning with Dextroscope in skull base tumor surgery, the VR-planned group had shorter operative time and postoperative length of stay, fewer intraoperative cerebrovascular injuries, and higher Karnofsky Performance Scores at discharge and at 6 months compared with conventional planning (Yang et al., 2009). This supports the premise that MRI-based planning tools can improve outcomes when validated.

Our findings are broadly consistent with previous functional and structural research suggesting that the putamen has distinct subregions with unique functional roles. For instance, previous studies show that the anterior putamen has stronger connectivity with limbic structures such as the nucleus accumbens and amygdala, whereas the posterior putamen is more connected to motor-related regions like the cerebellum and sensorimotor cortex (Zhang et al., 2016; Liu et al., 2018). Studies by Choi et al. (2012) using resting-state functional connectivity MRT (fcMRI) characterized this functional organization, which aligns with findings in primate studies. Additionally, Di Martino et al. (2008) analyzed the functional connectivity of basal ganglia circuitry in humans, highlighting distinct connections between striatal regions and cortical motor, cognitive, and affective regions, providing another in vivo evidence of a functional organization described in animals. More recent work used functional connectivity gradients and “gradientography” in a large healthy cohort to derive a multiscale subcortical atlas, including a dorsoposterior putamen subdivision (PUT-DP) with relatively high group-averaged Dice coefficients and stable coupling to sensorimotor cortical networks (Tian et al., 2020). In this context, our superior posterior segment is coarser and defined by stereotactic geometry rather than by gradients, but it occupies a similar dorsoposterior location and shows a comparable bias toward motor-associated connections, while other quadrants show stronger non-motor connectivity. Taken together, these comparisons suggest that an AC-based quadrant scheme, although simple, can approximate the location of putaminal territories described in more detailed functional and multimodal parcellations.

The choice of an AC-based segmentation was motivated by its compatibility with established stereotactic practice. The anterior commissure (AC), together with the posterior commissure (PC), is routinely used to establish the AC-PC line (Starr et al., 2006; Ostrem et al., 2007). Although newer techniques propose using alternative markers, such as the putamen itself, for deep brain stimulation (DBS) targeting (Thompson et al., 2017), the AC remains a reliable and easily identifiable structure across imaging modalities. Furthermore, algorithms have been developed for automatic AC detection, aiding segmentation and registration (Ardekani and Bachman, 2009), reinforcing the AC’s role in functional neurosurgery. While newer techniques may offer promising alternatives, particularly in research settings, the use of the AC-based division remains highly practical in clinical environments where time, resources, and the experience of neurosurgeons with emerging tools can be limited. In our data, this simple segmentation was sufficient to highlight a superior posterior region with predominantly motor-associated connectivity, which may be helpful when selecting a putaminal target for CED in PD.

The simulation algorithm was used to explore whether the superior posterior motor-dominant segment could be reached safely and with a quantifiable degree of simulated target-segment coverage via frontal and occipital injection trajectories. Trajectories were planned under a 2 mm safety margin relative to sulci, ventricles and vasculature. In most hemispheres, at least one safe trajectory could be identified for each entry type. In one patient with marked atrophy, no safe frontal trajectory could be planned, while an occipital trajectory remained feasible. This suggests that anatomical variability can lead to asymmetries in feasibility between entries, and that both options may need to be considered at the planning stage. Within each planned trajectory, a small number of spherical infusion volumes, constrained to remain within the superior posterior segment and respect safety margins, were positioned using a genetic algorithm to maximize overlap with the target. Under these conservative assumptions, frontal and occipital trajectories achieved similar target-segment coverage (<10% on average), with no significant difference between them. The relatively low coverage values should be interpreted in the context of the modelling choices. First, coverage was calculated with respect to the entire volume of the superior posterior segment, which includes tissue that may be difficult to reach with a single straight trajectory because of the curved and tapering shape of the segment. Second, infusion volumes were required to be fully contained within the segment and to observe safety margins, which reduces the number and size of spheres that can be placed along a given trajectory. Third, the simulations were limited to a small number of trajectories and spheres in order to remain close to what might be realistically implemented in a clinical setting. Finally, the diffusion of the infused agent was approximated as isotropic and spherical, without modelling tissue anisotropy or catheter-related backflow. Under these constraints, the <10% coverage values are better regarded as conservative estimates and as relative measures for comparing candidate trajectories, rather than as estimates of maximal achievable coverage. Higher coverage might be possible with additional entry points, longer in-segment path lengths, different catheter designs, or more detailed biophysical models, but such approaches were beyond the scope of this study.

Nevertheless, our simulation allowed us to refine injection site placement and volume along each trajectory to achieve optimal target coverage within safe anatomical limits. Using an adaptive approach, the algorithm could adjust placement configurations in response to coverage feedback, supporting the need for flexible trajectory planning when anatomical differences make one approach preferable. This approach could be useful in deciding which trajectory to choose for maximal possible coverage in cases where multiple trajectories are feasible.

Convection-enhanced delivery (CED) shows potential for conditions like PD, though it comes with risks, such as mass effect from infused drugs and neurotoxicity at high drug concentrations (Slevin et al., 2006; Taylor et al., 2013). A targeted approach focusing on motor-associated regions of the putamen could, in principle, help limit exposure of non-motor regions. However, whether this translates into better clinical outcomes or altered dosing requirements cannot be answered by the present simulations. Such questions require empirical studies that measure in vivo distribution and relate delivery metrics to outcomes. Nevertheless, such an approach aligns with recommendations in previous research on structural connectivity-guided interventions, such as DBS of globus pallidus internus, for improved outcomes in PD (da Silva et al., 2017). The simulations presented here illustrate one way in which connectivity information and basic geometric modelling can be combined to estimate, in advance, how different trajectories might perform with respect to a chosen putaminal segment.

Several extensions could be considered in future work. On the connectivity side, using higher-resolution diffusion imaging and models that account for crossing fibers could refine the estimation of putaminal connectivity patterns and test whether the posterior–superior motor bias observed here remains stable under different modelling assumptions. Normalized connectivity metrics, such as relative streamline densities adjusted for seed volume and total streamlines, could reduce the influence of ROI size and inter-individual differences in brain size. On the functional side, task-based or resting-state fMRI could be used to confirm that the structurally defined target segment participates in motor networks in individual patients. With regards to segmentation, separate cortical normalization could be implemented to further improve the alignment between the atlas and patient-specific anatomy, reducing the manual workload even further. Likewise, future studies with larger datasets might incorporate advanced techniques, such as deep learning, to better simulate drug distribution, although this approach requires significant data and computational resources. On the biophysical side, incorporating tissue anisotropy, regional permeability, or more complex infusion profiles (such as ramped-rate protocols) into the simulations could help relate modelled coverage more directly to in vivo drug distribution.

This study also has several limitations that must be addressed. The sample size was relatively small and limited to patients with advanced PD undergoing DBS work-up, which may restrict how broadly our findings may apply to a wider PD population or other patient groups. The diffusion protocol used a single b-value and standard clinical resolution, and deterministic, tensor-based tractography was applied without explicit modelling of crossing fibers, which may reduce sensitivity in regions with complex fiber architecture. The putamen segmentation was intentionally simple, based on AC-PC geometry rather than subject-specific functional gradients, and therefore aggregates finer subregions within each quadrant. Additionally, while our semi-automated cortical segmentation method reduced manual workload, it still required manual adjustments, introducing variability. The infusion simulations used a spherical, isotropic diffusion model and were not validated against empirical infusion data. Future studies that include measurement of in vivo distribution and clinical outcomes will be required to validate whether this planning approach improves delivery, safety, or patient-relevant endpoints.

This study used an AC-based four-quadrant segmentation and deterministic, clinically available tractography to characterize connectivity patterns within the putamen in patients with Parkinson’s disease and to explore their use for CED planning. The superior posterior segment showed the highest connectivity strength to motor-associated cortical and subcortical regions and was selected as the CED target. Using stereotactic planning and a simple geometric infusion model, both frontal and occipital drug injection trajectories into this segment could be planned safely in most hemispheres and yielded similarly comparable simulated target-segment coverage, with anatomical variation occasionally favoring one approach over the other. These findings suggest that connectivity mapping, even when derived from a coarse, AC-based segmentation and deterministic tractography, can help identify a motor-dominated putaminal segment and compare candidate trajectories for CED into this region. Further work with larger samples, crossing-fiber models, multimodal connectivity, more realistic diffusion models, and validation of infusion patterns will be needed to refine this approach.