Optimized Transnasal Electrical Stimulation Using High-Density Electrode Arrangements
DOI: Pending DOI
Synopsis
Recent work introduced transnasal electrical stimulation (TnES), a technique to modulate deep brain activity through the sphenoid sinus, a natural cavity in the skull base that is accessed routinely in transnasal procedures. TnES has the potential to provide minimally invasive treatment of neurological and neuropsychiatric conditions. We present a technique for optimizing the location, injected current amplitude, and number of electrodes to precisely and steerably target deep brain regions in TnES. Using rigorous simulations, we observe a 44.5% improvement in focality and 52.2% increase in intensity when targeting Brodmann Area 25, an emerging target for invasive DBS for depression. Further, we found that just 7 spatially-optimized electrodes retain 85% of the focality improvement.
Background
Deep Brain Stimulation (DBS) is an established treatment method for a wide range of chronic neurological disorders [1], and is an emerging treatment for treatment-resistant depression [2,3]. Despite the efficacy of DBS, the prerequisite of a highly invasive deep-brain surgery significantly reduces the population of patients willing and/or able to receive the treatment [4,5]. Non-invasive methods e.g. transcranial electrical/magnetic stimulation are often limited in their ability to reach the deep brain without substantial side effects [6,7].
TnES was recently proposed as a middle-ground between invasive deep-brain implants and noninvasive scalp contacts where electrodes are surgically placed on the walls of the sphenoid sinus and are used to stimulate deep brain regions such as Brodmann Area 25 (BA25) [8]. Guo et al. created the DeepFocus framework (DF), which leverages an MRI-based head model to simulate and optimize the electric fields generated by transnasally injected currents. Though this method exhibited optimized currents more focal and intense than scalp-only electrode configurations, it was applied to a fixed, sparse set of 10 sphenoid sinus-implanted electrodes. Since the location and number of electrodes in each configuration are critical treatment parameters which may affect stimulation safety, efficacy, tolerability, side-effects, and feasibility, this paper takes a step toward mitigating these key clinical barriers by answering these questions in-silica.
Methods
To best compare with the state-of-the-art results of Guo et al. [8], we use the same averaged head model – the New York Head volume conductor model [9] – to create a 0.5mm (voxel) resolution mask with tissue-type segmentations corresponding to white matter, gray matter, olfactory nerves, CSF, bone, skin, and air [7] (Fig 1.1). We leveraged this head model to create a novel methodology for an optimized placement of an approximately uniform, dense coverage of transnasal electrodes in the sinus. We integrated this into the DeepROAST pipeline using the Forssell et al. design [10] as a basis for our electrode volume and generated a dense spread of 91 electrodes (Fig 1.3). Using a segmented mask of BA25 as our target region (Fig 1.2), we then ran DF’s current optimization pipeline and compared the resulting electric fields created by the our dense 91 electrode configuration with that of DF’s sparse configuration according to Guo et al.’s optimization algorithm [8] (Fig 1.4). We used these optimization results to refine the original dense electrode configuration into a subset of the most useful electrodes. We performed an identical follow-up analysis on similar subsets, prioritizing electrodes with the largest absolute value injected currents, to analyze the benefit of precision electrode placement at lower density.
Results
We evaluated the dense, 91-electrode configuration alongside the sparse, 10-electrode configuration as a benchmark on focality, intensity, and off target activation. To ensure a robust comparison between the two frameworks, we evaluated their performance across a set of simulation states: we looked across 20 representative voxels within BA25 and 12 representative orientations. We then calculated the optimized maximum field intensity resulting from the DeepROAST pipeline at each target location and orientation with the default safety parameters — limiting the total injected current by 1mA.
We observe that the average electric field attainable with the dense electrode configuration outperformed that of the sparse configuration [8]: the dense configuration created a 1.84V/m average electric field, compared to 1.37V/m with the sparse configuration. We also found that the dense configuration resulted in an average 52.2% increase in maximum intensity.
Similarly to intensity, we computed the focality of each configuration using the same metric as the DeepFocus paper, defined as the volume of voxels over the region of interest (VVER). We ran simulation states across all locations and orientations where the desired electric field of our target voxel needed to exceed 10%, 30%, 50%, 70%, and 90% of the sparse framework’s max intensity. The dense arrangement provided a stark improvement in focality when compared to the sparse: the dense and sparse frameworks resulted in average VVERs of 14657mm^3 and 27708mm^3 respectively. Overall, the dense arrangement resulted in a 44.5% reduction in VVER on average. A brain-slice view of the electric field resulting from the dense and sparse configurations optimized for the target to achieve 0.123V/m is shown in Figure 1.4b and Figure 1.4a, respectively.
In addition to the metrics of intensity and focality in the DeepFocus work [8], we introduced a new metric, similar to focality, called cancel region activation (CRA). This metric is calculated by taking the sum of the magnitude of the electric field of all voxels considered outside the region of interest. Though this metric behaves similarly to VVER, it exhibits greater stability than VVER as it has no thresholding behavior. Because variability in neural dynamics and network effects can cause certain neurons to be activated even with current amplitudes that are deemed to be subthreshold, this metric provides greater insight into the aggregate activity of untargeted brain matter, which may lead to potential side effects of the stimulation. Similar to VVER, we observed an average decrease in CRA of 38.5% with dense arrangement.
We also leveraged this metric to have a more robust evaluation of the tradeoff of density. We performed a similar element-wise focality-optimized simulation where each configuration was modeled as a subset of the electrodes in the dense framework. Electrodes were ranked based on the absolute value of injected current after performing the DeepFocus max-focality optimization; electrodes with the largest injected currents received the highest ranking. For each simulation state, we evaluated the performance of subsets of 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 30, 40, 50, and 70 electrodes in addition to the fully dense 91 configuration. On average, it required only 7 electrodes to retain 85% of the reduction in CRA provided by the full dense configuration. A brain-slice view of the electric field resulting from 7 optimally-placed electrodes is shown in Figure 1.4c. With configurations using a number of electrodes greater than 7, we observe diminishing returns in the improvement of focality.
Discussion
Overall, our results show that our novel method of optimizing currents with dense electrode placements substantially outperforms that of DF's sparse configuration across stimulation intensity, focality, and off-target activation. This is expected to result in more effective, more tolerable, and farther-reaching TnES with fewer side effects, enhancing the effectiveness of TnES and further increasing the accessibility of TnES-based treatments. We answer the question of the benefit of dense electrode configurations with respect to stimulation focality and propose a novel methodology for the automated placement and choosing the ideal number of electrodes.
This simulation-based work has several limitations. All evaluations of stimulation patterns present in this work assume quasi-static time dynamics and focus on electric field optimization, which does not account for neural responses nor patient pain resulting from these fields. This limitation also means that waveform design remains an open research question. And while the injected current limits of DF may minimize the probability of lasting damage, further experimental validation is needed.
Additionally, the brain matter distribution and sphenoid sinus shape are variable across individuals. This means that the effectiveness of the automated electrode placement methodology may vary on a subject-by-subject basis. Thus, subject-specific CT scans should be utilized, and experimental results are needed to validate the effectiveness and tolerability of this modality.
Disclosures
KSH was funded through the NIH T32 Neural Interface Trainee program and the GEM Fellowship program with Synopsys as the designated employer. This work was supported by The Assistant Secretary of Defense for Health Affairs endorsed by the Department of Defense, in the amount of ($258 606), through the Peer Reviewed Medical Research Program under Award No. HT9425-24-1-0018. Opinions, interpretations, conclusions, and recommendations are those of the author(s) and are not necessarily endorsed by The Assistant Secretary of Defense for Health Affairs endorsed by the Department of Defense. YG was supported by the Centre for Machine Learning and Health (CMLH) fellowship at Carnegie Mellon University. This material is also based upon work supported by the Chuck Noll Foundation for Brain Injury Research.
References
[1] Thomas Wichmann and Mahlon R DeLong. Deep brain stimulation for neurologic and
neuropsychiatric disorders. Neuron, 52(1):197–204, 2006
[2] Wu, J Mo, L Sui, J Zhang, W Hu, C Zhang, and Xie. Deep brain stimulation in treatment- resistant depression: a systematic review and meta-analysis on efficacy and safety. Frontiers in neuroscience, 15, 2021.
[3] Helen S Mayberg, Andres M Lozano, Valerie Voon, Heather E McNeely, David Seminowicz,
Clement Hamani, Jason M Schwalb, and Sidney H Kennedy. Deep brain stimulation for
treatment-resistant depression. Neuron, 45(5):651–660, 2005.
[4] Albert J Fenoy and Richard K Simpson. Risks of common complications in deep brain stimulation surgery: management and avoidance. Journal of neurosurgery, 120(1):132–139, 2014.
[5] Robert F Dallapiazza, Philippe De Vloo, Anton Fomenko, Darrin J Lee, Clement Hamani, Renato P Munhoz, Mojgan Hodaie, Andres M Lozano, Alfonso Fasano, and Suneil K Kalia. Considerations for patient and target selection in deep brain stimulation surgery for parkinson’s disease. Exon Publications, pages 145–160, 2018.
[6] Sven Bestmann and Vincent Walsh. Transcranial electrical stimulation. Current Biology, 27(23):R1258–R1262, 2017.
[7] Ji Hyun Lee, Young Cheol Yoon, Hyun Su Kim, Jiyeong Lee, Eunju Kim, Christian Findeklee, and Ulrich Katscher. In vivo electrical conductivity measurement of muscle, cartilage, and peripheral nerve around knee joint using MR-electrical properties tomography. Scientific Reports, 12, 2022.
[8] Yuxin Guo, Mats Forssell, Dorian M Kusyk, Vishal Jain, Isaac Swink, Owen Corcoran, Yuhyun Lee, Chaitanya Goswami, Alexander C Whiting, Boyle C Cheng, et al. Deepfocus: A transnasal approach for optimized deep brain stimulation of reward circuit nodes. Journal of Neural Engineering, 22(1):016048, 202
[9] Yu Huang, Abhishek Datta, Marom Bikson, and Lucas C Parra. Realistic volumetric-approach to simulate transcranial electric stimulation—roast—a fully automated open-source pipeline. Journal of neural engineering, 16(5):056006, 2019.
[10] Forssell, M., Guo, Y., Kusyk, D., Cheng, B., Whiting, A., Wang, E. W., & Grover, P. (2026). Transnasal implants in the nasal cavity and sphenoid sinus for minimally invasive deep brain stimulation: measurement of electric field in cadavers. IEEE Transactions on Biomedical Engineering.