Each patient in our center underwent extensive evaluation by our pediatric neurology and neurosurgery teams before pursuing neurosurgical intervention. This includes discussion at our weekly Epilepsy Surgery Conference, consisting of a panel of neurosurgeons, epileptologists, neuroradiologists, neurologists, advanced practice providers, and neuropsychologists. Evaluation included assessment of each patient's MRI, video EEG (vEEG), functional MRI, diagnostic intracranial EEG with either cortical grids/strips or stereo-EEG (sEEG), magnetoencephalography (MEG), and positron emission tomography (PET), as indicated.

Seizure semiology was determined using a combination of patients' clinical features and seizure localization on inpatient vEEG. DRE was defined as failure to control seizures using adequate trials of at least two ASMs (3). Data collected on implantation technique included prior sEEG or cortical grid/strip monitoring, device implanted, and subsequent neurosurgical interventions, among other variables. For response to neuromodulation, each patient was placed into 1 of 5 categorical outcomes: (1) free of disabling seizures; (2) ≥90% reduction of disabling seizures; (3) ≥50% but <90% reduction of disabling seizures; (4) <50% but some noticeable reduction of disabling seizures; or (5) no improvement (30–33). Outcomes were only considered for patients who had a follow up of at least 1 year or more to allow for sufficient time for neuromodulation to have an effect. Post-surgical complications, revisions, explants, and device side effects were also noted.

A total of 17 patients underwent device implantation, 9 with DBS and 8 with RNS. The demographic and clinical characteristics are summarized in Table 1. The age at epilepsy onset was 4.2 ± 1.3 years for DBS patients (mean ± SE). For RNS patients, age at epilepsy onset was 8.0 ± 1.5 years. The seizure semiology differed between groups, with generalized seizures being more common in DBS patients (78%) compared to RNS patients (12%). DBS patients more frequently presented with more than one seizure type as compared to RNS patients and had greater frequency of daily or weekly seizures. Details on seizure burden and frequency for both groups are detailed in Table 2.

All patients had failed at least two ASMs and were classified as having DRE. The mean number of ASMs at time of presurgical evaluation was similar between the groups (DBS: 3.7 ± 0.5; RNS: 2.8 ± 0.3). DBS patients had prior trials of ketogenic diet (56%) and VNS therapy (67%). Patients who had prior VNS therapy had VNS for 4.8 ± 0.6 years (range 2.4–7.3) with stimulation parameter adjustment before undergoing DBS placement. No patients trialed these therapies in the RNS cohort.

Four DBS patients (44%) had prior neurosurgical interventions aimed at controlling seizures, while no RNS patients had prior neurosurgical interventions. Among the four DBS patients, two patients underwent callosotomy, one patient underwent focal resection, and one patient underwent placement and revision of a ventriculoperitoneal shunt. Patient 2 with Lennox-Gastaut Syndrome (LGS) underwent a 2/3 corpus callosotomy which was partially helpful in decreasing seizure burden; however, she continued to have seizures after surgery and had resultant left-sided weakness and supplementary motor area syndrome. Patient 4, also with LGS, first underwent an anterior 2/3 callosotomy, and then later underwent an additional craniotomy for ablation of the splenium of the corpus callosum. Patient 6 with underwent a right anterior temporal lobectomy as well as right frontopolar resection.

Device selection and surgical targeting were tailored to the patients' specific clinical profiles, including epilepsy etiology, seizure semiology, trialed therapies (ASMs, ketogenic diet, VNS), and prior neurosurgical interventions (Table 3). DBS was more often used for used for generalized epilepsy, particularly in patients with LGS or other generalized epilepsy syndromes (78%), targeting the centromedian nucleus (CMN) in most cases (78%). In contrast, RNS was employed mostly in patients with focal or multifocal epilepsy (88%), guided by invasive monitoring such as sEEG or grid/strip intracranial recordings. The rationale for device selection often reflected a balance between seizure control and preserving functional cortex, with RNS prioritized when seizure foci overlapped eloquent cortex (e.g., motor or somatosensory regions, see Patients 11 and 13).

All surgeries were performed under general anesthesia. sEEG monitoring with ROSA was utilized in 11% of DBS patients compared to 100% of RNS patients (Table 2). The mean age at the time of neuromodulation device placement was 16.8 ± 0.9 years for DBS patients and 15.3 ± 1.7 years for RNS patients.

The most common DBS target location was the CMN (78%), followed by the ANT (11%), and the pulvinar nucleus of the thalamus (11%). All RNS patients underwent sEEG recording which was used to determine placement location, most often a seizure onset focus or foci (63%), followed by the bilateral hippocampi (25%) or bilateral temporal lobes (12%) (Figures 1, 2). Identified seizure onset foci were the in the left anterior parietal region, the right motor cortex, the canthus/middle temporal gyrus, and the right insula and right frontal operculum.

Postoperative complications occurred two DBS patients. Patient 1 experienced a prolonged 14-day postoperative recovery, characterized by waxing and waning mental status, respiratory distress requiring BIPAP, prolonged fevers, and labile blood pressures and fluid status (Table 3). Contributing factors likely included stress from pneumonia, underlying poor reserves, ICU delirium, and hyperammonemia encephalopathy, with seizures managed through adjustments in ASMs. Patient 4 developed swelling and redness along her DBS site and chest incision after implant of the pulse generator. This later became infected, requiring readmission, return to the operating room, surgical exploration, and explant 6 weeks after initial implant. The patient was treated with 2 weeks of IV antibiotics. No RNS patients had postoperative complications.

Revision surgery was necessary in one patient with RNS (13%) and one with DBS (11%). Patient 14 underwent a repositioning and replacement of RNS leads in conjunction with a resection of the left temporal pole and amygdala 1.6 years after initial placement. Patient 1 had their DBS pulse generator implanted deep to the pectoral muscle initially which led to difficulty charging the device. The patient underwent revision surgery 11 months after initial surgery to have the pulse generator placed superficial to the muscle, which resolved the issue with charging.

Three RNS patients underwent additional neurosurgical interventions during or following RNS placement. Patient 11 underwent resection of a left parietal epileptic focus during the same surgery as RNS placement. Patient 13 underwent three resective surgeries after RNS placement: the first consisted of resections around the motor and sensory areas, the second was resection of a right frontal epileptic focus, and the third involved a resection near the motor strip, which resulted in an infected subgaleal fluid collection requiring drainage.

In Patient 14, post-implantation RNS was actively used to map seizures and guide decisions regarding future resections. Despite ongoing adjustments to RNS settings, the patient continued to experience daily seizures. Over multiple clinic visits, RNS data consistently localized seizure onset to the left mesial temporal lobe, specifically the hippocampus, where the hippocampal lead showed significant ictal activity compared to a lateral strip electrode. This information guided the decision to consider hippocampal transection as a memory-preserving surgical option given the patient's high verbal memory function. This was favored over laser ablation, which was more likely to negatively impact memory. After this transection, the patient continued to have cluster seizures prompting a second craniotomy for left amygdalectomy and temporal pole resection with RNS lead repositioning. After this repositioning of RNS leads, the RNS system continued to capture seizures, supporting that the epileptic tissue had not been fully resected. The patient underwent a third resection surgery consisting of a temporal lobectomy and replacement of RNS leads. As of the time of publication, three years after initial RNS placement, the patient has achieved clinical seizure freedom.

Only one DBS patient had a side effect of worsened symptoms of depression noted two years after initial implant. This resolved once the patient began seeing a counselor. No RNS patients had documented side effects.

Among the 12 patients with a follow up of one year or more (7 DBS, 5 RNS), the mean age at last follow up was 20.5 ± 1.3 years for DBS patients and 22.1 ± 1.4 years for RNS patients. The mean duration of follow-up was 4.0 ± 0.4 years for DBS patients and 4.8 ± 0.8 years for RNS patients. At this time, the mean number of ASMs was 3.9 ± 0.6 for DBS patients and 2.0 ± 0.4 for RNS patients. ASM burden decreased in 14% of DBS patients and 50% of RNS patients.

All patients had reduced seizure burden relative to pre-neuromodulation. Of the RNS patients with at least a 1 year follow up, 100% (n = 5) had a >90% reduction in seizure burden, and 80% achieved a seizure freedom. Of the DBS patients with at least a 1 year follow up, 71% of DBS patients had a ≥50% reduction in seizures and 28% had a <50% reduction. Among the DBS patients with CMN as a target (n = 6), 67% percent had a ≥50% reduction in seizure burden and 33% had a <50% reduction. No patients experienced a lack of effect or worsening of seizures with either device-based therapy. The change in seizure burden after therapy is presented in Table 4.

A multifactorial approach that considers a patient's prior epilepsy history, neuroimaging findings, genetic testing, and ASM response is necessary to decide between neuromodulation vs. resection or ablation. Resection surgery is the first choice when a focal lesion is clearly identified. Larger resections or hemispherectomy may be necessary when multifocal or unilateral diffuse lesions are present, although this poses challenges when eloquent cortex is present (e.g., bitemporal, primary motor/sensory, or language areas) (22, 34). Neuromodulation may be a better option in patients who have complex, bilateral epileptogenic networks, such as those with LGS (35). In patients with LGS, DBS has been shown to reduce seizure burden and also improve attention, behavioral issues, and quality of life (36, 37). Lesions in eloquent cortex are also an indication for possible neuromodulation. Patients with average or above average language and memory function may experience significant postoperative impairments if regions such as the hippocampus are resected or ablated. This was the case for several patients who received RNS in our study. Finally, neuromodulation can serve as a palliative option when prior drug therapies and surgical interventions resection or have failed, as was the case for 44% of DBS patients in our study.

In this cohort, DBS was typically used as a palliative procedure for patients with generalized epilepsy (78%) who had undergone multiple prior neurosurgical interventions (44%) and were also refractory to VNS (67%) and ketogenic diet therapy (56%). In addition, DBS patients who received VNS had a mean therapy duration of 4.8 ± 0.6 years (range 2.4–7.3) before DBS. Of note, the DBS patients in this cohort appeared had more severe epilepsy than the RNS patients, with 100% of DBS patients presenting with GTC seizures vs. 40% of RNS patients, an earlier mean age of seizure onset in DBS patients (4.2 vs. 8.0 years), and more DBS patients resistant to prior neurosurgical, dietary, and VNS interventions. Prior work, including a partially randomized trial, has shown that DBS as an add-on to VNS therapy is more beneficial than VNS with continued parameter tuning alone (36, 38, 39). The potential synergistic effects of DBS and VNS are not well understood and may involve the modulation of multiple networks (38).

Of the DBS patients who had a least a one-year follow up, 71% had a ≥50% reduction in seizures as of last follow up. This is similar to that of a large review by Khan et al., where 75% of patients had at least a ≥50% seizure reduction (19). Most patients in our study had electrodes targeting the CMN, and 67% of these patients experienced a ≥50% reduction in seizure burden. The CMN was shown to be a promising target in the prospective, double-blind randomized ESTEL trial (40). Other potential targets, such as the ANT, may also be further considered, as it is currently approved for treating focal epilepsy in adults (6, 18). This was the target of one DBS patient in our cohort who demonstrated a >90% seizure reduction at last FU.

RNS is particularly useful for patients with focal or multifocal epilepsy where the stimulating electrode can be placed into or on top of the cortical seizure onset zone, even when it overlaps with eloquent cortex. For our RNS patients, the majority had focal or multifocal epilepsy (88%) with an older age of seizure onset (mean 8.0 ± 4.2 years), and none had prior failed neurosurgical interventions. As performed for all RNS patients in our study, sEEG and grid/strip intracranial monitoring can provide precise seizure localization to optimize placement of the RNS electrode. However, modulation of the seizure onset zone may not be sufficient for patients that have more diffuse epileptogenic networks, as is seen in many of those with extratemporal lobe epilepsy (41). More research is needed on how to identify whether a focus-driven or network-driven approach will be more effective in a given patient.

Among our RNS patients who had at least 1 year follow up, 80% achieved seizure freedom and 100% achieved a ≥ 50% reduction in seizure frequency as of last follow up. This data is consistent with prior reports of RNS in pediatric DRE. Swartwood et al. reported a cohort of 22 patients, 64% of whom demonstrated a ≥50% reduction in seizure frequency at a mean follow up of 19 months (31). Similarly, in a multi-center study by Nagahama et al., 59% of 17 pediatric DRE patients achieved a ≥50% seizure reduction (30).

Future research should focus on identifying patient selection criteria for RNS and DBS, optimization of the timeliness of device placement, evaluating long-term seizure and quality-of-life outcomes, and developing innovative strategies to map and treat diffuse epileptic networks. Current clinical decision-making often prioritizes identifying candidates for focal resection or callosotomy based on seizure type, with neuromodulation considered as a subsequent or palliative procedure only after other interventions fail. However, there is a need to shift toward a more strategic, biomarker-driven approach that identifies patients who are not ideal resective surgical candidates from the outset. Understanding when a patient's epileptic network is too large or complex for resection could allow for earlier selection for neuromodulation as the primary intervention, sparing patients a series of invasive and ineffective procedures. The impact of neuromodulation on quality of life as compared to other approaches warrants further consideration. Moreover, future studies should focus on defining the clinical and neurophysiological markers that can guide preoperative decision making.

Recent research efforts have highlighted the potential of individual differences in the functional and structural connectivity of thalamocortical networks as biomarkers for optimizing patient selection, stimulation parameters, and neuromodulation targets (43–45). Local field potentials, which may be recorded with newer DBS devices, may also be helpful in optimizing programming settings or providing closed-loop stimulation (46, 47). Integrating machine learning and artificial intelligence into neuromodulation devices may enable dynamic seizure prediction, or aid in the optimization of individualized stimulation parameters in the future. Currently underutilized or unidentified clinical or neuroimaging biomarkers may prove important, indicating a need for further research associated with patient selection and improving outcomes.

Limitations of this study include the cohort size, the potential selection bias of these patients, that this is a single-institution study, and that some patients underwent other epilepsy surgeries that may have had a contributory effect to the results presented here. Nonetheless, our study demonstrates the safety and efficacy of DBS and RNS in a relatively clinically diverse cohort of patients with long-term follow up in most patients.