Method and system for predicting deep brain stimulation parameters

Abstract

The present disclosure relates to a method for suggesting personalized and symptom-specific DBS stimulation parameters, comprising the steps of: creating a three-dimensional symptom network library comprising a data-driven three-dimensional model of symptom-specific nerve tracts in a stereotactic space established based on a detailed and comprehensive pathway profile; registering patient data with the three-dimensional symptom network library to analyze how an electrode of a single patient is mapped thereto; stimulation parameters are derived and suggested from baseline symptom severity characteristics for each patient. Another object of the invention relates to a system for predicting DBS parameters, the system comprising an interface for providing DBS parameters to an electrode and receiving signals from a patient, and a computer programmable unit comprising a data memory for storing a network library of three-dimensional symptoms, the three-dimensional symptom network library includes a stored data driven three-dimensional model of symptom-specific nerve tracts in a stereotactic space established based on detailed and comprehensive pathways for predicting DBS parameters.

Description

Technical Field

[0001] This disclosure relates to a method and system for predicting deep brain stimulation (DBS) parameters. Background Technology

[0002] Deep brain stimulation (DBS) is a well-established treatment for Parkinson's disease. Over the past decade, DBS research has undergone a paradigm shift, moving towards investigating the effects of DBS on distributed brain networks (Horn et al., 2017; Horn and Fox, 2020). Identifying and characterizing such networks in a symptom-specific manner could pave the way for personalized DBS.

[0003] Deep brain stimulation (DBS) of the subthalamic nucleus (STN) is a well-established treatment for Parkinson's disease (PD). Since the introduction of levodopa, it has been a pivotal historical advancement, fundamentally changing the way the disease is treated. However, although randomized clinical trials have demonstrated the efficacy of DBS in treating symptoms such as tremor and bradykinesia, its effectiveness remains to be seen. 1However, its effects on gait and other axial symptoms vary, and in some cases, electrical stimulation can even have harmful effects (Schrader, C. et al., *Neurology*, Vol. 77, pp. 483-488, 2011; Yin, Z. et al., *Journal of Neuroscience*, Vol. 393, pp. 116-127, 2018). This is further supported by the fact that 200,000 patients worldwide have received DBS treatment (Veda-Mai, V. et al., *Frontiers in Human Neuroscience*, Vol. 15, 2021), but according to large clinical trials, approximately 71,000 patients (35%) did not experience significant improvement in quality of life despite relief of motor symptoms (Deuschl, G. et al., *The New England Journal of Medicine*, Vol. 355, pp. 896-908, 2006). Therefore, while many patients benefit from DBS, not all do (Aviles-Olmos, I. et al., *Journal of Neurology, Neurosurgery & Psychiatry*, Vol. 85, pp. 1419-1425, 2014). One reason may be that current clinical practice treats all symptoms of the disease by surgically targeting a focal brain region. For example, in STN-DBS, the target coordinate is located in the posterolateral part of the subthalamic nucleus, a location determined by direct imaging and surgical landmarks such as the Bejjani line (Bejjani, B.-P. et al., *Journal of Neurosurgery*, Vol. 92, pp. 615-625, 2000). Although stimulation parameters can be adjusted during DBS programming postoperatively, and heuristics exist for optimizing symptom-specific parameters in clinical practice (e.g., switching to dorsal contacts to treat tremor), a more comprehensive, data-driven model for optimizing patient-specific symptoms is still lacking. Furthermore, the segmented electrodes currently used contain up to 16 contacts per lead, further complicating the programming process.

[0004] Therefore, precisely defining symptom-specific network targets appears to be one approach to improve clinical responses to DBS. Evidence for symptom-specific networks dates back to the 1960s, when meticulous analysis of lesion studies revealed networks causally linked to specific core symptoms of Parkinson's disease (Hassler et al., *Brain*, Vol. 83, pp. 337-350, 1960; McGregor, MM and Nelson, AB, *Neuron*, Vol. 101, pp. 1042-1056, 2019). For example, the Freiburg stereotactic school, based on pioneering research of 560 ablation cases between 1950 and 1958, concluded by Hassler et al. that optimal control of tremor required the destruction of the circuit between the cerebellum (and Morales' triangle), which sends axonal collaterals to the red nucleus on its way to the ventrolateral posterior nucleus (Vop) of the thalamus and the primary motor cortex (Hassler et al., *Brain*, Vol. 83, pp. 337-350, 1960). In contrast, connections between the globus pallidus and the ventrolateral anterior nucleus of the thalamus (Voa) and the supplementary motor area (defined as area 6a⍺ by the Vogt / Hassler / Brodmann school) are associated with improvements in symptoms of reduced movement, such as bradykinesia and rigidity. Years later, modern neuroimaging studies confirmed these two views. For example, Helmich et al. found that resting tremor in Parkinson's disease (and likely includes action tremor) is associated with the cerebellar-thalamic-cortical circuit (Ni et al., Annals of Neurology, Vol. 68, pp. 816-824, 2010; Helmich et al., Contemporary Neurology and Neuroscience Research Reports, Vol. 13, p. 378, 2013; Sturman et al., Brain, Vol. 127, pp. 2131-2143, 2004). Researchers such as Akram et al. have also confirmed this through DBS network mapping. In addition, they found that the improvement of bradykinesia and rigidity symptoms is associated with connections from the premotor cortex and prefrontal cortex (Akram et al., Neuroimaging, Vol. 158, pp. 332-345, 2017; Strotzer, QD et al., Annals of Neurology, Vol. 85, pp. 852-864, 2019).

[0005] Given the fine structure of the dissociated but related overlapping circuits involved, optimal efficacy of DBS for Parkinson's disease can be achieved by treating each symptom or symptom cluster (rather than the entire disease) as an independent entity that can be treated by modulating specific circuits. This leads to two conclusions: First, an accurate three-dimensional symptom-circuit model needs to be established in stereotactic standard space. Once established, patient-specific electrode implantation sites can be correlated with this model to determine the optimal stimulation parameters for each patient: if the patient presents with tremor-dominant symptoms, parameters can be adjusted to maximize modulation of the tremor circuit; conversely, if axial symptoms have the greatest impact on quality of life, parameters can be adjusted to maximize the effect on the axial circuit. Second, by activating different contacts (or combinations of contacts)—and possibly by applying stimulation pulses of different frequencies—it is possible to treat multiple dissociated circuits using a single DBS electrode. This complicates parameter selection and highlights the need for methods that automatically suggest symptom-specific stimulation parameters.

[0006] Publicly available US Patent 10,905,882B2 discloses a system and method for optimizing DBS pulse signal parameters for patient treatment. In predicting optimal DBS parameters, functional brain data is input into a prediction system. This functional brain data is obtained by scanning in a multidimensional parameter space of one or more DBS parameters. The prediction system extracts statistical indices of brain responses from functional brain data of one or more regions of interest (ROIs) or voxels in the brain and accesses a DBS functional atlas containing disease-specific brain response maps derived from DBS treatment under optimal DBS parameter settings for multiple diseases or neurological disorders. Based on the statistical indices of brain responses and the DBS functional atlas, the prediction system predicts one or more optimal DBS parameters for the patient.

[0007] Publicly available US Patent 11,395,920B2 relates to a system and method for identifying patient-specific neurosurgical target locations. The system receives patient brain imaging data containing neural bundles and networks in the patient's brain, accesses a quantitative connectivity atlas comprising population-based disease-specific structural and functional connectivity maps, the connectivity maps containing neural bundle and network patterns associated with optimal target regions (OTAs) identified from the patient population, and defines patient-specific neurosurgical target locations based on a comparison between neural bundle and network patterns in the patient brain imaging data and OTA-related neural bundle and network patterns identified from the patient population in the quantitative connectivity atlas. This quantitative connectivity atlas is a population-based disease-specific quantitative connectivity atlas that identifies optimal therapeutic target locations associated with maximum clinical improvement for each disease in the patient population.

[0008] Publicly available US Patent 8,792,991B2 describes techniques for modeling treatment fields for therapy provided by a medical device. Each treatment field model is based on a set of treatment parameters and represents the range of treatment propagation of a treatment system providing treatment according to that set of parameters. The treatment field model can be used to guide modifications to the treatment parameters. For example, a processor compares an algorithmic model of the treatment field with a reference treatment field and adjusts at least one treatment parameter based on that comparison. As another example, the processor adjusts at least one treatment parameter to improve the operating efficiency of the treatment system while substantially maintaining the modeled treatment field.

[0009] Publicly available U.S. patent application US2011 / 040351A discloses a computer-implemented method comprising storing a tissue activation volume (VTA) data structure from multiple patient analyses. The method involves receiving patient data for a specific patient, representing an assessment of the patient's condition. The VTA data structure is then evaluated relative to the patient data to determine a target VTA for achieving the desired therapeutic effect for that specific patient. Summary of the Invention

[0010] The first aspect of this disclosure relates to a method for suggesting personalized and symptom-specific DBS stimulation parameters, comprising the steps of: creating a three-dimensional symptom network library containing data-driven three-dimensional models of symptom-specific neural bundles in stereotactic space based on detailed and comprehensive pathway maps; registering patient data with the three-dimensional symptom network library to analyze how electrodes for individual patients are mapped to it; and deriving and suggesting stimulation parameters based on the baseline symptom severity characteristics of each patient.

[0011] Another object of the present invention relates to a system for predicting DBS parameters, the system comprising an interface for providing DBS parameters to electrodes and receiving signals from a patient; and a computer-programmable unit comprising a data storage for storing a three-dimensional symptom network library, the three-dimensional symptom network library comprising a stored data-driven three-dimensional model based on detailed and comprehensive pathway establishment of symptom-specific neural bundles in stereotactic space for predicting DBS parameters.

[0012] Other aspects, features, and advantages of the invention will readily become apparent from the following detailed description, which only describes preferred embodiments and implementations. The invention can also be implemented in other different embodiments, and several details thereof may be modified in various obvious respects without departing from the spirit and scope of the invention. Therefore, the drawings and description should be considered illustrative rather than restrictive in nature. Additional objects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0013] The invention will now be described with reference to the accompanying drawings. It should be understood that the embodiments and aspects of the invention described in the drawings are merely examples and are in no way intended to limit the scope of the claims. The invention is defined by the claims and their equivalents. It should be understood that a feature of one aspect or embodiment of the invention may be combined with features of other aspects or embodiments of the invention, wherein: Figure 1 Exemplary electrode implantation sites are shown, in which the active contacts of each center in the discovery queue, the validation queue I, the validation queue II, and the prospective queue are visualized.

[0014] Figure 2 A shows the symptom-specific nerve bundles in a sagittal view, magnified at the subthalamic nucleus (STN) level.

[0015] Figure 2 B shows symptom-specific nerve bundles separately at the STN level, while other nerve bundles are shown in gray for spatial contrast.

[0016] Figure 2 C shows symptom dissociation within the indirect pathway connection between the STN and the globus pallidus, following a similar head-tail-frontal gradient.

[0017] Figure 2 D indicates the cortical origin of superdirect projection.

[0018] Figure 3 A shows that the tremor nerve tract includes projections from the cerebellar nucleus (to the thalamus) and cortical projections from the primary motor cortex (to the STN).

[0019] Figure 3 B shows the neural bundles associated with axial symptoms, including brainstem connections to the pontine peduncle (PPN) region.

[0020] Figure 3 C shows that after separating axial symptoms into gait-related items and all other items, the connection is found to be driven by gait (rather than other axial symptoms).

[0021] Figure 3 D shows a comparison of projection sites with matching sections from a histological atlas (published by Coulombe et al. in Frontiers in Neuroanatomy, 2021).

[0022] Figure 4 A shows the stimulation field and symptom-specific nerve bundles of the same patient.

[0023] Figure 4 B shows the stimulation volume of the example patient and the optimal nerve bundle associated with improvement in the Unified Overall Parkinson's Disease Rating Scale-III (UPDRS-III).

[0024] Figure 5 A shows the symptom-specific neural bundles found in the cohort that were associated with improvement in Parkinson's disease symptoms.

[0025] Figure 5 B shows the symptom-specific neural bundles associated with improvement in Parkinson's disease symptoms in validation cohort I (41 cases in Würzburg, 53 cases in Beijing).

[0026] Figure 5 C shows a correlation plot between the empirical improvement in validation queue I and the predicted improvement (predicted by the discovery queue).

[0027] Figure 6 A shows a rain cloud plot, where each data point represents the Spearman correlation coefficient between the estimated and empirical UPDRS-III improvement in a setup of one of 20 electrodes. The one-sided t-test results are significant, indicating that the Spearman correlation coefficient (rho) is positive for most patients (T=4.15, p=5.3e-04, mean R=0.41±0.44). Red ellipses represent the stimulation contacts that yielded the greatest improvement in a particular patient; model-selected contacts are marked with blue ellipses, and the corresponding stimulation fields are shown for example electrodes.

[0028] Figure 6 B shows the symptom specificity of the model: repetition Figure 6 The analysis in A was performed, but this time the bradykinesia symptom was weighted to the greatest extent. When the model was used to estimate bradykinesia improvement, the correlations were almost all positive across all 20 electrode settings, but the correlations dropped significantly when used to predict stiffness improvement. The paired t-test between the two models was significant (T=3.2987, p=0.0045).

[0029] Figure 6 C shows the symptom specificity of the model: repetition Figure 6 The analysis in A was performed, but this time the rigidity symptom was weighted to the greatest extent. When the model was used to estimate rigidity improvement, the correlations were almost all positive across all 20 electrode settings, but the correlations dropped significantly when used to predict bradykinesia improvement. The paired t-test between the two models was significant (T=2.5484, p=0.02).

[0030] Figure 7 The comparison between stimulation parameters generated by conventional treatment standard settings and stimulation parameters generated according to the method of this disclosure is shown.

[0031] Figure 8 A shows a well-positioned standard omnidirectional electrode (Medtronic 3389) whose single stimulation volume can uniformly cover all symptom-specific nerve bundles.

[0032] Figure 8 B illustrates a hypothetical future conception using modern electrodes.

[0033] Figure 9 A shows an example fiber bundle (dashed line) in the pathway map.

[0034] Figure 9 B shows that this process is repeated for all fiber bundles in the pathway map to create a symptom network library.

[0035] Figure 9 C shows a single nerve tract model (encoding overall motor improvement) that is cross-validated by estimating motor improvement based on activation of the nerve tract in left-out patients.

[0036] Figure 9 D shows a more complex symptom-specific neural bundle model that repeats the process four times for each symptom neural bundle and weights the estimates based on the baseline intensity of each symptom.

[0037] Figure 10 An example of treating a new (hypothetical) patient according to the method of this disclosure is shown.

[0038] Figure 11 The statistical results of 1000 10-fold cross-validations are shown. Left panel: Comparison of the correlation value distribution of the 10-fold cross-validation for the four-branch model and the single-branch model. Middle panel: Correlation between the empirical improvement and predicted improvement of the four-branch model. Right panel: Correlation between the empirical improvement and predicted UPDRS-III value improvement of the single-branch model.

[0039] Figure 12 A shows a visualization of the combined bradykinesia and rigidity of the neural bundles.

[0040] Figure 12 B shows repeated analysis after regression from stiffness improvement to bradykinesia improvement (and vice versa), showing no qualitative changes at the anatomical level. Detailed Implementation

[0041] The technical problem is solved by the independent claims. The dependent claims cover other specific embodiments of the invention.

[0042] The term “stimulation parameters” refers to the number of pulses, pulse frequency, pulse width, pulse train frequency, amplitude, application of combined pulses through multiple electrodes, and the implantation location of at least one electrode for each pulse train.

[0043] The term "symptom score" refers to a subset of items from a validated, reproducible scoring system used to assess disease severity and treatment response. For example, items in the motor section of the Unified Parkinson's Disease Rating Scale-III (UPDRS-III) that measure tremor severity, frequency, etc., constitute a tremor "symptom score." Changes in these items before and after surgery (or with DBS on and off) are associated with improvement in the symptom score. Within the scope of this disclosure, "symptom score" applies to symptoms presenting in movement disorders.

[0044] This disclosure provides a method and system aimed at making progress by: i) establishing a three-dimensional loop model of disease-specific symptoms; ii) utilizing the model in a method for suggesting optimal stimulation parameters, said stimulation parameters being a function of the baseline symptom severity characteristics of each patient.

[0045] The model has been further subdivided, for example, into four symptom categories (tremor, bradykinesia, rigidity, gait and other axial symptoms) (see [link to relevant documentation]). Figure 3 Furthermore, this concept can be easily applied to other conditions, such as obsessive-compulsive disorder, whose symptoms have been subdivided into obsessive thoughts, compulsive behaviors, anxiety, depression, cognitive flexibility, and overall functioning.

[0046] A model was developed according to this disclosure and cross-validated on data from three patient cohorts in the discovery cohort (N=43, Würzburg; N=35, Amsterdam; N=51, Berlin; see Table 1), two patient cohorts in validation cohort I, one patient cohort in validation cohort II, and five prospective patients. All patients in the discovery cohort received bilateral STN-DBS for Parkinson's disease using a four-contact omnidirectional electrode (Medtronic 3389), with bilateral hemispheric activation stimulation. Electrodes in all 129 patients were located in the thalamic base region, with active contacts located in this region (…). Figure 1 Clinical scores for all patients showed that the mean baseline score of the Unified Parkinson's Disease Rating Scale (UPDRS)-III was 44.59 ± 14.30 (standard deviation), with a mean improvement rate of 51.71 ± 24.26%. The Würzburg cohort analysis used the MDS-UPDRS, while the Amsterdam and Berlin cohorts studied the conventional UPDRS (see Table 1).

[0047] Table 1. Demographic Information To create the symptom network library of this disclosure, an extended version of the DBS fiber tractography atlas (Middlebrooks, EH et al., *American Journal of Neuroradiology*, Vol. 41, pp. 1558-1568, 2020) was used to define the anatomical connections to and from the STN (see Methods), which was validated through extensive literature review and, where possible, through Klingler anatomy in collaboration with professional neuroanatomists. Using the DBS fiber filtering method (Baldermann, JC et al., *Biological Psychiatry*, Vol. 85, pp. 735-743, 2019), the study investigated which nerve bundles included in the pathway atlas were associated with improvements in bradykinesia, rigidity, tremor, and axial symptoms (symptom network library). The symptom network library can take into account other nerve bundles associated with other symptoms and can therefore be designed to be indication-specific.

[0048] DBS fiber filtering is a univariate method that yields statistical coefficients (e.g., Spearman rank correlation coefficients in this case) for each neural bundle connected to a stimulus volume group. Multiple comparison correction was performed on the correlation coefficients using the Benjamin-Hohberg method (false discovery rate) at an α level of 0.05, retaining only the surviving significant neural bundles. This group of significant fibers showed a clear head-to-tail gradient of symptom improvement at the thalamic base level. Figure 2 Importantly, the same gradient was reflected throughout the network, indicating a “symptom localization” arrangement in both the superdirect pathway and the globus pallidus-thalamus projection. The head-to-tail order of the indirect projections (from the globus pallidus to the STN) was consistent with the findings of the aforementioned superdirect / cortical-thalamus studies. Figure 2 C).

[0049] The nerve tracts associated with tremor improvement project from the primary motor cortex to the posterior region of the motor STN. As expected, the tremor tracts also include the crossed cerebellar-thalamic pathway. The nerve tracts associated with rigidity improvement project from the premotor supplementary motor area (SMA) and similar premotor frontal areas to the anterior part of the premotor thalamus. Between the tremor and rigidity tracts, the nerve tracts associated with bradykinesia and axial symptoms improvement overlap anteroposteriorly. However, the bradykinesia tracts enter from the inner surface of the STN, while the axial tracts terminate in their lateral portions (see [link to article]). Figure 2 (Illustration in A) – Both originate from the SMA and adjacent lateral cortical regions. Axial tracts also include restricted connections to the midpontine peduncle (PPN) region of the brainstem.

[0050] Figure 2 A symptom network database is shown according to this disclosure. Figure 2A shows the symptom-specific nerve bundles in a sagittal view, magnified at the STN level. The symptom-specific nerve bundles follow a head-tail-frontal gradient, with the tremor bundle being the most posteriorly positioned, followed by the bradykinesia, axial symptoms, and rigidity bundles. All bundles shown were significant after multiple comparison correction (p<0.05). It should be noted that these bundles are adjacent to each other, thus all (or most) of the bundles can be modulated with a single well-placed electrode (consistent with clinical experience that a single stimulation field can often provide some relief for many / all motor symptoms).

[0051] See Figure 2 B. Symptom-specific nerve bundles are visualized individually at the STN level, with other nerve bundles shown in gray for spatial contrast. The inset shows the results of 10-fold cross-validation and each symptom-specific nerve bundle. Figure 2 C shows symptom dissociation within the indirect pathway connection between the STN and the globus pallidus, following a similar head-tail-frontal gradient. Figure 2 D shows the cortical origin of superdirect projection. The nerve bundles associated with improvement in tremor originate from the primary motor cortex, while the nerve bundles associated with improvement in bradykinesia symptoms originate from the premotor cortex in a more dispersed manner.

[0052] As mentioned above, Figure 2 All nerve bundles shown were significant after multiple comparison correction. The robustness of the model was further tested. For this purpose, permutation analysis was first performed on symptom-specific nerve bundles. Here, except for the tremor nerve bundle and the axial nerve bundle, the variance explained by all nerve bundles was significantly greater than that explained by the nerve bundle model after recalculating the patient's improvement value 1000 times (p<0.05). Secondly, k-fold cross-validation was performed on the nerve bundle model. Again, except for the tremor nerve bundle, all nerve bundles explained a significant amount of variance in 10-fold cross-validation (bradykinesia: R=0.20, p=0.02; rigidity: R=0.20, p=0.02; axial symptoms: R=0.22, p=0.01, see also). Figure 2 Crucially, although all symptoms except tremor were computable across the entire cohort, the tremor analysis included only 29 of the 129 patients because 100 patients had a baseline tremor score below 3 or a 100% improvement rate (this explains why the tremor model, despite its significance, was less robust than other tracts when subjected to permutation or cross-validation tests). The tremor analysis had to be limited in this way because patients without significant tremor at baseline would certainly not have achieved significant improvement with DBS (even with optimal stimulation), while patients with a 100% improvement rate might have achieved further improvement if they had more tremor at baseline. Given the factor structure and disease characteristics of UPDRS, this problem does not exist for other symptoms (see also Methods).

[0053] Figure 3 Anatomical validation of the symptom network database is shown. For example... Figure 3 As shown in A, the streamlines associated with tremor improvement include the cerebellar-thalamic pathway. This is combined with precise cortical projections originating from the primary motor cortex (…). Figure 2 D), these precise connections have been widely considered to be associated with tremor in a large body of literature (Hassler et al., *Brain*, Vol. 83, pp. 337-350, 1960; Akram, H. et al., *Neuroimaging*, Vol. 158, pp. 332-345, 2017; Coenen, VA et al., *Acta Neurosurgica Sinica (Vienna)*, Vol. 18, pp. 130-14, 2020; Helmich et al., *Annals of Neurology*, Vol. 69, pp. 269-281, 2011; Helmich et al., *Journal of Brain Neurology*, Vol. 135, pp. 3206-3226, 2012). Similarly, neural tracts associated with improved axiality include brainstem connections restricted to the PPN nucleus region ( Figure 3 B). The pontine parietal nucleus (PPN) has been considered a promising stimulation target for treating gait problems (part of the axial symptom group), although efficacy varies (Mazzone et al., *Neurosurgery*, Vol. 73, p. 894, 2013; Zrinzo, L. et al., *Brain*, Vol. 131, pp. 1588-1598, 2008). Given this clinical relevance, these connections were tested to determine whether they are specific for gait improvement (or related to all axial symptoms). Neural bundles associated with gait and axial symptoms include connections between the STN and the pontine parietal nuclei. Upon separating gait-specific symptoms from axial symptoms, it was found that gait items drive this STN-PPN connection (…). Figure 3 (C and D). Since axial symptoms (especially gait problems) are by far the most difficult to treat with DBS (Klawans, HL, *Movement Disorders*, Official Journal, Movement Disorders Society, Vol. 1, pp. 187–192, 1986; Bonnet et al., *Neurology*, Vol. 37, pp. 1539–1542, 1987), this finding is of particular value. Furthermore, these results again align with known associations between anatomy and motor function (i.e., the specific association between gait improvement and PPN regions), which further qualitatively validates the findings of this study.

[0054] Previous studies have shown that bradykinesia and rigidity share a common neural basis (as opposed to tremor) (Kühn, AA et al., *Experimental Neurology*, Vol. 215, pp. 380-387, 2009), which does not directly match the degree of separation between the rigidity and bradykinesia neural bundles found in this study. To further investigate this, regression analysis was performed on improvements in rigidity and vice versa, yielding the same separation results. Figure 12 ).

[0055] although Figure 2 All results shown were statistically significant after multiple comparison correction, but further investigation was conducted to determine whether the entire model (rather than each individual symptom bundle) could robustly estimate overall motor improvement when cross-validated. Figure 4 A illustrates this analysis. For this purpose, patients were randomly assigned to one of ten folds. The four-branch model was then iteratively recomputed, leaving one fold uncomputed each time. The stimulation sites (represented by the magnitude of the electric field vector, i.e., the E-field) of the uncomputed patients were overlaid with the fascicle model and multiplied by the weighted optimal fascicle assigned to each fascicle (each fascicle represented by the Spearman rank correlation coefficient) to derive fiber scores for each stimulation field and each symptom. These estimates were then weighted based on each patient's preoperative symptom score (i.e., for patients with severe tremor at baseline, high fiber scores from the tremor fascicle were given greater weight). This resulted in a single combined fiber score, which was mapped to (predicted) UPDRS-III improvement by a linear model computed on the training set (i.e., nine out of ten folds). Crucially, the model did not see any data from the estimated uncomputed folds in each iteration (see Methods for details). These predicted improvements were significantly correlated with empirically observed relative improvements in UPDRS-III in patients (R=0.33, p=0.0001, mean absolute error: 17.87%±14.1%). Figure 4 A).

[0056] The same analysis was repeated after calculating the percentage improvement in overall UPDRS-III for individual neural tracts that directly encode the data. Figure 4 As shown in B, this model is simpler. Instead of weighting the fiber scores from each symptom tract, a single fiber score is calculated for each patient, directly encoding the percentage improvement in UPDRS-III. This single-tract model is more likely to mimic previous studies aimed at determining the optimal structural connectivity features for overall motor improvement (Horn, A. et al., Annals of Neurology, Vol. 82, pp. 67-78, 2017; Treu, S. et al., Neuroimaging, Vol. 219, pp. 117018, 2020). Compared directly with the four-symptom model, the single-tract model performed worse (R=0.28, p=0.001, mean absolute error: 18.11% ± 14.3%). Figure 4 B). When tested over multiple iterations of the randomized folds, the correlation based on the multi-branch model was significantly higher than that based on the single-branch model (T=93.7, p=2e-16).

[0057] Finally, we ensured that the choice of k (in k-fold cross-validation) would not significantly affect the results. Therefore, we repeated the analysis using 5-fold and 7-fold cross-validation, and similar significant results were obtained again for both the multi-branch model (R=0.37, p=0.0002 and R=0.36, p=0.0002, respectively) and the single-branch model (R=0.35, p=0.0004 and R=0.27, p=0.002, respectively).

[0058] In the second validation step, our model was used to estimate clinical outcomes from an out-of-sample dataset of 93 retrospective Parkinson's disease patients from two different centers (Würzburg = 52 cases, Beijing = 41 cases). The symptom-specific model trained on the discovery cohort significantly predicted the outcomes of validation cohort I (R = 0.37, p = 0.0006). Qualitatively, the pathways associated with Parkinson's disease symptoms in validation cohort I were similar to those in the discovery cohort. Specifically, the connection between M1 and the STN, and the cerebellar tract, were associated with tremor improvement. Improvement in axial symptoms was associated with streamlines immediately anterior to the vertebra, followed by streamlines associated with stiffness improvement (SMA and prefrontal cortex regions).

[0059] To take a first step toward clinical application, a method according to this disclosure has been developed that maximizes stimulation of any symptom-specific nerve bundle for new patients. The method according to this disclosure employs an alternative optimizer and suggests stimulation settings, then predicts outcomes in a manner similar to the multi-bundle predictive analysis described above. Taking into account the individual patient's baseline motor score, the method optimizes contact location by maximizing the overlap between the stimulation volume and the relevant symptom-specific nerve bundle. Therefore, for example, patients with high tremor severity and severe axial damage, the same electrode implantation location will result in different settings. As a patient's symptom characteristics change over many years, the method according to this disclosure can be used to update the suggested parameters during the patient's illness.

[0060] To validate the method according to this disclosure, it was first applied to all patients in an out-of-sample validation cohort I (retrospective cohort), suggesting optimal parameters based on each patient's baseline symptom characteristics. The proposed settings were then compared with actual clinical parameters. Post-hoc tests showed that patients suggested for the same contact point according to the method of this disclosure showed significantly greater clinical improvement than patients suggested for different contact points (R=0.23, p=0.03). Crucially, due to the retrospective nature of the cohort, the actual clinical improvement suggested by the parameters proposed according to the method of this disclosure cannot be obtained; therefore, this analysis relies on the effectiveness of the predictive model. In other words, the outcome predictions of the parameters proposed according to the method of this disclosure were compared with the outcome predictions of the clinical parameters.

[0061] Given the success of retrospective validation of this method, an attempt was made to apply its findings to predict symptom specificity in 10 patients undergoing unipolar re-examination (validation cohort II, Cologne). Specifically, these 10 patients had repeat data points, encompassing the outcome (or until the onset of side effects) associated with each contact stimulation at a range of amplitudes from 1 mA to 5 mA. Thus, a total of 186 clinical outcomes were obtained across the 10 patients. First, the overall UPDRS improvement in validation cohort II was predicted using the method according to this disclosure, with positive predictions for all but four electrodes. Therefore, the one-sample t-test for R values ​​at the 20 electrodes was positive (T = 4.155, p = 5.3e-04). For each stimulation setting, data on bradykinesia and rigidity improvement were obtained separately. To test symptom specificity, the analysis was repeated twice, each time with bradykinesia or rigidity weighted to the maximum extent in a multi-branch model. For the model weighted for the correct symptoms, the correlation between the estimated values ​​and empirical improvement was significantly higher for each electrode setting than for the corresponding other symptoms (p<0.05 for both analyses). Figure 6 (B and C).

[0062] To test the feasibility of the method according to this disclosure in clinical application, the DBS stimulation parameters recommended according to the method of this disclosure were tested in five patients. Figure 7 Electrode positioning and two stimulation protocols (this disclosure vs. standard treatment; SoC) and their tract overlap from a multi-tract model are shown. In summary, the baseline UPDRS-III score was 49.8 ± 22.1 points, representing an improvement of 34.4 ± 13.1 points (73 ± 11.8%) under the method settings according to this disclosure, and an improvement of 31.8 ± 15.1 points (65.4 ± 12.1%) under the standard treatment settings. In four of the five patients, the improvement achieved by the method settings according to this disclosure was greater than that of the standard treatment settings; in the fifth patient, the improvement was comparable (36 points vs. 38 points). Although three of the five patients preferred the method settings according to this disclosure over the standard treatment settings, these settings resulted in side effects in two patients (patient 05 experienced motor dysfunction, and patient 04 experienced dizziness). This indicates that the current model is purely based on improvement (rather than side effects), which is a clear limitation for clinical application. Future attempts should include adding tracts encoding side effects to the model. Figure 7 The comparison between stimulation parameters generated by conventional treatment standard settings and stimulation parameters generated according to the method of this disclosure is shown.

[0063] This disclosure is the first to propose a three-dimensional model of symptom-specific neural tracts in a stereotactic space, built in a data-driven manner based on detailed and comprehensive pathway maps. This disclosure demonstrates the robustness of this symptom network library to cross-validation. For example, potential applications have been shown on single-tract models based on overall UPDRS-III improvement calculations. Based on the generated model, a method capable of suggesting personalized and symptom-specific DBS stimulation parameters was created in 129 patients and preliminarily validated in a discovery cohort and two different out-of-sample datasets from five different centers (validation cohort I, N=93; validation cohort II, N=10). Finally, the DBS settings suggested according to the method of this disclosure were prospectively applied in five patients. Compared to the clinical settings, motor symptoms improved in four patients.

[0064] The results presented in this disclosure support the view that, in Parkinson's disease examples, different networks are associated with improvements in core symptom categories. Specifically, symptom-dissociated basal ganglia-thalamus-cortical motor circuits can be isolated via symptom-specific pathways arranged caudate-to-caudate within the sensorimotor-premotor functional areas of the STN. Although dispersed at the cortical level, each symptom-specific tract originates primarily from different cortical regions. Furthermore, tremor connectivity includes connections from the cerebellar nuclei, while axial symptoms include connections with brainstem PPN regions.

[0065] It needs to be clarified that these results do not indicate that a symptom domain can be independently modulated by a specific neural tract. There is a considerable degree of overlap between connections, particularly at the cortical level and indirect (globus pallidus-thalamic base) projections. On the other hand, the projection areas of ultradirect (cortical) input to the STN appear quite separated—although very close to each other. At first glance, this may seem to contradict clinical experience: in fact, the same DBS setup can modulate multiple symptoms simultaneously, and the intensities appear to be similar. However, this view does not conflict with the presented results: it must be emphasized that the identified neural tracts are very close to each other, spanning areas several millimeters within the sensorimotor functional areas at the STN level. Figure 8 As shown in Figure A, a single well-positioned electrode can generate a stimulation volume that simultaneously modulates all identified nerve bundles (and thus modulates symptoms). However, Figure 8 B illustrates the potential application of a nerve tract model with modern 16-contact segmented electrodes (such as the Boston Scientific CartesiaX model). Using multi-independent current control techniques, different stimulation volumes, each with different amplitudes and frequencies, can be activated along the same electrode (Timmermann, L. et al., *The Lancet Neurology*, Vol. 14, pp. 693-701, 2015). Figure 8In the hypothetical example shown, a high-frequency (180Hz) volume can be directed to the tremor nerve bundle, and another low-frequency (25Hz) volume can be directed to the axial and gait nerve bundles to treat both symptoms as optimally as possible.

[0066] Therefore, the results of this disclosure may have clinical relevance: First, segmented electrodes can focus stimulation volumes with increasing precision. This leads to a surge in parameter space, making imaging-guided methods indispensable (Roediger, J. et al., Movement Disorders, Vol. 37, pp. 574-584, 2022; Roediger, J. et al., The Lancet Digital Health, Vol. 5, e59-e70, 2023). As imaging methods and electrode localization become increasingly precise, this disclosure provides a method for fine-tuning stimulation parameters according to each patient's symptom needs using a symptom network library (such as the symptom network library in this disclosure). This will facilitate prospective validation of the results of this study (Roediger, J. et al., The Lancet Digital Health, Vol. 5, e59-e70, 2023; Gadot, R. et al., Biological Psychiatry, 2023). Second, although the stimulation site of a single well-positioned electrode can cover most of the identified nerve bundles, the location of the stimulation peak may still be important (and not all lead implantation sites are optimal).

[0067] To further develop this approach, a proof-of-concept process was introduced to obtain recommendations for symptom-specific DBS parameters. This process allows for the setting of weights for symptoms that are prevalent and burdensome in patients, and the tuning of parameters to maximize stimulation of the symptom-specific network. For example, this process can be run for typical tremor-dominant patients (high tremor intensity, low bradykinesia-rigidity symptom intensity), which would facilitate a setting that maximizes targeting of tremor connectivity from the primary motor cortex and cerebellum.

[0068] This disclosure proposes, firstly, defining symptom-specific neural bundles at the population level using standard connectivity data. Next, patient data is spatially registered with the resulting symptom-branch library to analyze how electrodes for individual patients map to it. Then, the method is practically personalized at the symptom level using a concept previously known as “network fusion” (Hollunder, B. et al., Progress in Neurobiology, 102211, 2021; Holllunder et al., Biological Psychiatry: Cognitive Neuroscience & Neuroimaging, Vol. 6, 939-941, 2021). This concept involves fusing (or weighting) identified symptom networks to derive optimal stimulation targets for symptom features prevalent in an individual patient. Brain sensing combined with machine learning can provide immediate feedback to the DBS system, automatically instructing the process to switch network targets according to the patient’s current individual needs, such as when symptoms like tremor occur under stress (Merk, T. et al., Experimental Neurology, Vol. 351, 113993, 2022). One day, this could open up new horizons for the integration of adaptive DBS technology with symptom-specific connectomics, leading to a real-time, personalized, and precision medicine DBS approach.

[0069] In summary, this disclosure provides a method for a three-dimensional loop model based on disease-specific symptoms, and its use in proposing stimulation parameters that incorporate a characteristic function of the severity of a patient's baseline symptoms. Furthermore, this disclosure introduces a method capable of using this model to propose DBS-programmed symptom-specific stimulation parameters, which ultimately improves clinical outcomes and patient satisfaction.

[0070] To demonstrate the effectiveness of the methods according to this disclosure, a retrospective study was conducted on 129 patients with Parkinson's disease (PD) who received STN-DBS treatment (discovery cohort). Of these, 51 were treated in Berlin, 43 in Würzburg, and 34 in Amsterdam (patient characteristics and demographic data are shown in Table 3).

[0071] Table 3: Statistics of Clinical Scores and Sub-item Scores for Each Cohort All patients underwent bilateral implantation of two quadrupole DBS electrodes (model 3389; Medtronic, Minneapolis, Minnesota). The study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of Charité University School of Medicine. Furthermore, this method prospectively applied DBS settings to five Parkinson's disease patients who underwent DBS surgery at Würzburg University Hospital based on a retrospective database. This application was approved by the Institutional Review Board of the University of Würzburg.

[0072] The percentage improvement in the motor component of the Unified Parkinson's Disease Rating Scale-III (UPDRS-III) was calculated as an indicator of overall treatment outcome, based on the difference between pre- and post-operative scores divided by the pre-operative score. For the Würzburg cohort, the revised MDS-UPDRS-III was used. Similarly, improvements in UPDRS-III items representing the four major motor symptoms of Parkinson's disease were calculated: bradykinesia (items 23, 24, 25, 26, measuring finger tapping, hand movement, rapid alternating movement, and leg flexibility [and available MDS item toe tapping]), rigidity (item 22, measuring stiffness in the neck, arms, and legs), tremor (UPDRS items 20 and 21 [and available MDS tremor items: MDS postural tremor, MDS kinetic tremor, MDS resting tremor lip / jaw, and resting tremor persistence]), and axial symptoms (items 27, 28, 29, 30, measuring posture, postural stability, and gait [and available MDS item frozen gait]). Gait scores (items 29 and 30) were further listed separately in the sub-analyses, along with combinations of all axial symptoms excluding these gait items. In the analysis, the baseline tremor sub-score had a higher standard deviation (4.27 ± 5.62 points), while other sub-scores were relatively lower (Table 3). Patients with a baseline tremor score below 2 points and those with a relative improvement rate of 100% were excluded from the tremor analysis. Preoperative multispectral MRI scans were acquired in a routine clinical setting to define patient-specific anatomical targets. Postoperatively, patients underwent CT (N=84) or MRI scans (N=45) to locate the electrodes.

[0073] DBS electrodes were positioned using the Lead-DBS software (Horn et al., Neuroimaging, Vol. 184, pp. 293–316, 2019; Horn, A. and Kühn, AA., Neuroimaging, Vol. 107, pp. 127–135, 2015) according to the revised scheme of version 3 (Neudorfer et al., Neuroimaging, Vol. 268, pp. 119,862, 2023). In short, this involved linear registration of postoperative and preoperative images using advanced normalization tools (ANTs) (Avants et al., Medical Imaging Analysis, Vol. 12, pp. 26–41, 2008). The registered images were then normalized to the ICBM2009b nonlinear asymmetric (“MNI”) template space using the ANTsSyN method, employing an efficient low-variance + subcortical thinning scheme implemented in Lead-DBS (Ewert, S. et al., Neuroimaging, Vol. 184, pp. 586-598, 2019). The results of each preprocessing step were visually reviewed, and thinning was performed as necessary. In particular, the normalization error was corrected using the WarpDrive module available in Lead-DBS (Neudorfer, C. et al., Neuroimaging, Vol. 268, p. 119862, 2023). After preprocessing, the phantom-validated PaCER method (Husch et al., Clinical Neuroimaging, Vol. 17, pp. 80-89, 2018) was used for DBS electrode localization on postoperative CT, or the TRAC / CORE method or manual localization method was used on postoperative MRI data (Horn, A. and Kühn, AA, Neuroimaging, Vol. 107, pp. 127-135, 2015).

[0074] To estimate the stimulation volume, the magnitude of the electric field vector (E-field) around the electrodes was calculated. This was based on a four-chamber mesh that distinguishes between gray and white matter, electrode contacts, and insulating portions. The gray matter region was defined by the DISTAL map (Ewert, S. et al., Neuroimaging, Vol. 170, pp. 271–282, 2018). The static formulation of the Laplace equation was then solved on a discrete domain represented by a tetrahedral four-chamber mesh using an improved version of the FieldTrip-SimBio pipeline (Vorwerk et al., Biomedical Engineering Online, Vol. 17, p. 37, 2018).

[0075] Creating a comprehensive anatomical pathway map of subcortical regions: Multiple network mapping methods used data-driven connectomes derived from diffusion-weighted imaging-based fiber tractography (Baldermann et al., Biological Psychiatry, Vol. 85, pp. 735-743, 2019; Horn et al., Annals of Neurology, Vol. 82, pp. 67-78, 2017; Al-Fatly et al., Brain, Vol. 142, pp. 3086-3098, 2019; Li et al., Nature Communications, Vol. 11, p. 3364, 2020). However, the accuracy of these datasets is often insufficient, especially in the subthalamic region, particularly regarding small projection tracts such as the globus pallidus-thalamic projection (lentiform loop and lentiform tract), Eddinger's comb fibers, or the cerebellar-thalamic pathway (Middlebrooks et al., *American Journal of Neuroradiology*, Vol. 41, pp. 1558-1568, 2020; Petersen et al., *Neuron*, Vol. 104, pp. 1056-1064, e3, 2019; Noecker et al., *Journal of Neuromodulation*, Vol. 24, pp. 248-258, 2021; Alho et al., *Motion Disorders*, Vol. 35, pp. 75-80, 2020; Horn et al., *Neurology*, Vol. 92, e1663-e1664, 2019). To overcome this limitation, a key advance was the use of prior anatomical knowledge to define the concept of manually curated pathway atlases (Hamani et al., *Brain*, Vol. 127, pp. 4–20, 2004; Marani et al., *Advances in Anatomy, Embryology and Cell Biology*, Vol. 198, pp. 1–113, p. vii, 2008). These datasets contain carefully curated pathways and therefore have no false-positive connections (neural bundles not present in the brain). However, a drawback of these atlases is that neural bundles not defined by the anatomical team cannot be identified as key to successful DBS by using the atlas retrospectively. This can lead to false-negative conclusions (no neural bundles present in the brain were found). For example, the undoubtedly most accurate atlas defined by Petersen et al. lacks bundles such as the subthalamic loop (Petersen et al., *Neuron*, Vol. 104, pp. 1056–1064, e3, 2019), the connection between the STN and PPN, or the striatum-substantia nigra connection, simply because the anatomical team had to focus on a limited number of neural bundles and could not exhaustively include all neural bundles present in the brain. In addition, the STN receives cortical (superdirect) input from the entire frontal cortex (Yeh, Nature Communications, Vol. 13, p. 4933, 2022; Isaacs et al., Frontiers in Neuroanatomy, Vol. 12, 2018), which is not reflected in typical pathway maps.To overcome this limitation, this paper compiles existing atlases (Middlebrooks et al., American Journal of Neuroradiology, Vol. 41, pp. 1558-1568, 2020; Petersen et al., Neuron, Vol. 104, pp. 1056-1064.e3, 2019), including cortical and subcortical pathways across the thalamus, and adds missing connections based on anatomical data (Table 2).

[0076] Table 2 shows the nerve tracts defined in the DBS fiber tract imaging atlas, revised version 2. Missing connections not represented in any atlas were calculated using the same methodology used to create the first version of this atlas, as detailed in other literature (Middlebrooks, EH et al., *American Journal of Neuroradiology*, Vol. 107, p. 64, 2018). In summary, using a diffusion template from 1065 participants from the Human Connectome Project (VanEssen, DC et al., *Neuroimaging*, Vol. 62, pp. 2222–2231, 2012) (Yeh, F.-C. et al., *Neuroimaging*, Vol. 178, pp. 57–68, 2018), connections to the STN were mapped using seeds from various atlases (Table 2) (defined by the DISTAL atlas, Ewert, S. et al., *Neuroimaging*, Vol. 170, pp. 271–282, 2018). Therefore, this revised DBS fiber tract imaging atlas 2 (Middlebrooks, EH et al., American Journal of Neuroradiology, Vol. 41, pp. 1558-1568, 2020) contains a more comprehensive set of connections to and from the STN.

[0077] Multi-bundle implementation of DBS fiber filtering: This method is developed based on the concept of DBS fiber filtering (Baldermann, JC et al., *Biological Psychiatry*, Vol. 85, pp. 735-743, 2019) and extended in (Li, N. et al., *Nature Communications*, Vol. 11, p. 3364, 2020) to isolate neural bundles associated with changes in multiple motor symptom domains. Figure 8In the first step, this method is used to exemplarily construct a “symptom network library” that associates nerve bundles with improvements in clinical sub-scores for tremor, bradykinesia, rigidity, and axial symptoms. Activation of these four groups of nerve bundles is associated with improvements in the corresponding symptoms. To be included in this library, each nerve bundle must pass through a small number of E-fields (>0.5% of the total number of E-fields) with a reasonably high peak intensity (>1.5 V / mm). This constraint is set to exclude nerve bundles that are not strongly modulated by any stimulus field (theoretically, these nerve bundles might still obtain high correlation values ​​if subthreshold intensity is associated with clinical improvement). Changing any chosen values ​​(>0.5% of E-field and >1.5 V / mm), for example, to >3 and >4 V / mm, does not qualitatively change the results. Then, Spearman's rank correlation coefficient is calculated for each nerve bundle for each symptom group by correlating the individual sub-scores with the peak amplitude of the E-field for each patient through which the given streamline passes. This univariate approach generates a large number of rank correlation coefficients, and after multiple comparison correction using the false detection rate (FDR), its threshold is set to p-value < 0.05.

[0078] Clinical improvement was estimated using a symptom network library: the first 1000 fibers from the resulting fascicle group were used to estimate clinical improvement using a k-fold cross-validation design by overlapping the E-field of the patient with the corresponding symptom fascicle. Here, k (number of folds) was set to 10, as this is a standard choice in the field of machine learning. Different numbers (e.g., k=7 and 5) were tested, and similarly significant results were obtained.

[0079] Similarly, each pair of nerve tracts and E-fields is quantified by multiplying the peak intensity through which the nerve tract passes by the R-value assigned to that nerve tract (weighted for nerve tracts strongly correlated with improvement in a specific symptom). For each E-field, these values ​​are averaged to derive the mean fiber score for each symptom and E-field. The scores are then weighted against the baseline score of the predominant symptom for each patient. Thus, the modulation of tremor nerve tracts in each patient has a greater impact in patients with predominant tremor.

[0080] This yielded a combined fibrillation score encoding a fusion of patient-specific symptom improvements. The fibrillation score was further mapped to the percentage improvement for each symptom using a linear model applied to the corresponding training cohort. The predicted improvement for each symptom was then converted to a predicted percentage improvement in UPDRS using a weighted average, where each weight is a standardized baseline score for that symptom. To exclude the possibility of accidental significant results due to random patient assignment to the training and test sets, the k-10 cross-validation process was iterated 1000 times, with patients randomly assigned to the training and test sets in each iteration.

[0081] A method for proposing DBS programming parameters based on a symptom network library: This disclosure proposes a method for proposing DBS stimulus parameters based on a symptom network library. Figure 9This method uses an alternative optimizer to consider all possible unipolar DBS stimulation fields (within the range of a given electrode) and estimates the improvement for each hypothetical stimulation field in the same manner as described above. The predictions are then ranked to determine the most effective stimulation setting. As previously mentioned, nerve bundles can be weighted according to symptom severity, patient burden, or patient preference. For example, if a patient is most bothered by tremor or seeks a setting that suppresses tremor in a specific situation, a higher weight can be assigned to tremor improvement. The following table describes the algorithm in detail.

[0082] Formula 1 Cleartune algorithm 1. Set optimizer parameters, such as maximum number of samples, minimum number of random samples required to create alternative models, objective constraints, etc.

[0083] 2. Define the current boundary (see Formula 1) 3. Initialize J init (Initial current) 4. Solve for J init This is a FEM (Finite Element Method) problem. If necessary, solve for additional random samples within the current boundary to create an alternative model.

[0084] 5. Exploring the parameter space using alternative models a. Using a balanced parameter space to explore and minimize the value function of the alternative model, random sampling is performed around the current optimal solution (the best solution observed so far). The sample with the smallest value is the adaptive sample.

[0085] b. Solve the FEM problem for adaptive samples and improve the alternative model.

[0086] c. If a new global minimum (i.e., minimum error or maximum value) is observed, update the current optimal solution.

[0087] 6. Update the sampling dispersion based on the success rate of the adaptive sample relative to the current optimal solution.

[0088] 7. If convergence is achieved before the maximum number of iterations is reached, but the evaluation value is still lower than the target value, then reset by discarding all adaptive points.

[0089] The method was first tested by applying it to the same retrospective cohort, but a prospective trial was conducted with five patients to test its feasibility in a clinical setting. However, this method was not intended to test its non-inferiority or superiority over any existing method.

[0090] Another object of this disclosure is a system for predicting DBS parameters and providing predicted stimulation parameters to electrodes. The system includes a computer-programmable unit (CPU) with a data storage containing a three-dimensional symptom network library, which contains data-driven three-dimensional models of symptom-specific neural bundles in stereotactic space based on detailed and comprehensive pathway construction. The CPU is also configured to register patient data with the created three-dimensional symptom network library to analyze how electrodes for a single patient map to it, thereby deriving the optimal stimulation volume for the patient's symptom profile.

[0091] When the predicted DBS parameters meet the patient's needs, they can be applied to electrodes previously implanted in the patient's surgery via a system interface. It will be apparent to those skilled in the art that at least one electrode must be implanted in at least one region of the brain, and multiple electrodes may already be implanted. The electrodes may be segmented electrodes capable of stimulating multiple nerve bundles.

[0092] The system may further include a pulse generator for generating the pulses required for DBS. It is also conceivable that the pulse generator may have been pre-operatively implanted, so that the system's predictive parameters are transmitted to the pulse generator via an interface, and the pulse generator then transmits the pulses to at least one electrode.

[0093] The system may also include a sensor system capable of sensing and tracking the DBS signals transmitted by the system. The sensor system can also detect input signals from the electrodes and provide them to the CPU to acquire patient data that can be used to predict and optimize DBS parameters.

[0094] The foregoing description of preferred embodiments of the invention is presented for illustrative and descriptive purposes. It is not intended to be exhaustive or to limit the invention to the precise forms disclosed, and modifications and variations are possible in accordance with the foregoing teachings or the practice of the invention. These embodiments were chosen and described to explain the principles of the invention and its practical application, so that those skilled in the art can use the invention in various embodiments suitable for particular applications. The scope of the invention is intended to be defined by the appended claims and their equivalents.

Claims

1. A method for suggesting personalized and symptom-specific DBS stimulation parameters, comprising the following steps: - Create a three-dimensional symptom network library, which contains a data-driven three-dimensional model of symptom-specific nerve bundles in stereotactic space based on a detailed and comprehensive pathway map; - Register patient data with the three-dimensional symptom network database to analyze how electrodes for individual patients are mapped to it; - Based on the baseline symptom severity characteristics of each patient, derive and recommend stimulation parameters.

2. The method according to claim 1, wherein, Creating the three-dimensional symptom network library includes the following steps: - Specific nerve bundles for limiting symptoms; - Determine the correlation coefficient for each symptom-specific nerve bundle that associates stimulation volume with any change in symptom score; - Significant symptom-specific nerve bundles were identified by adjusting the correlation coefficient.

3. The method according to claim 2, wherein, The symptom-specific neural bundles involve pathways to and from the subthalamic nucleus.

4. The method according to claim 2 or 3, wherein, The stimulation volume was associated with improvements in bradykinesia, stiffness, tremor, and axial symptoms.

5. The method according to any one of claims 2 to 4, wherein, The correlation coefficient was corrected using the Benjamin Hochberg method via multiple comparisons.

6. The method according to any one of claims 1 to 5, wherein, The three-dimensional symptom network database includes a four-symptom network database, which covers four core motor symptom regions including tremor, bradykinesia, rigidity, and axial symptoms.

7. The method according to any one of claims 1 to 6, wherein, The three-dimensional symptom network database is validated before registering patient data with it.

8. The method according to claim 6, wherein, Stimulation parameters are recommended for each symptom-specific nerve bundle.

9. The method according to any one of claims 1 to 8, wherein, The baseline symptom score of an individual patient is considered part of the patient data.

10. The method according to any one of claims 1 to 9, wherein, Recommended stimulation parameters include maximizing the overlap between the stimulation volume and the corresponding symptom-specific nerve bundles.

11. The method according to any one of claims 1 to 10, wherein, Segmented stimulation electrodes are recommended for stimulation parameters.

12. The method according to any one of claims 1 to 11, wherein, Symptom weights are considered in the symptom characteristics of patient data to suggest stimulus parameters.

13. The method according to any one of claims 1 to 12, wherein, Symptom-specific nerve bundles were defined at the population level using standard connectivity data, and patient data were spatially registered with the resulting three-dimensional symptom nerve bundle library to analyze how electrodes for individual patients are mapped to it.

14. The method according to any one of claims 1 to 13, wherein, The stimulation parameters relate to the recommended electrode implantation location before surgery.

15. A system for predicting DBS parameters, the system comprising: - An interface for providing DBS parameters to the electrodes and receiving signals from the patient; as well as - A computer programmable unit including a data storage device for storing a three-dimensional symptom network library containing a three-dimensional model driven by stored data of symptom-specific neural bundles in stereotactic space based on detailed and comprehensive pathway establishment, for predicting DBS parameters.

16. The system according to claim 15, wherein, The computer-programmable unit is configured to register patient data with the three-dimensional symptom network database to analyze how electrodes for a single patient are mapped to it.

17. The system according to claim 16, wherein, The computer programmable unit is configured to derive and suggest stimulation parameters based on the baseline symptom severity characteristics of each patient.

18. The system according to any one of claims 15 to 17, comprising a pulse generator for generating pulses via connection to at least one electrode provided to the patient.

19. The system according to any one of claims 15 to 18, comprising a sensor system for sensing and tracking DBS signals and / or receiving signals from the patient.

20. A method for treating movement disorders, comprising the following steps: - Connect the system used to predict DBS parameters to at least one electrode surgically implanted in at least one region of the patient's brain; - Apply DBS parameters to at least one electrode of a patient with a movement disorder, wherein the DBS parameters are determined in advance by: creating a three-dimensional symptom network library containing a data-driven three-dimensional model of symptom-specific neural bundles in stereotactic space based on a detailed and comprehensive pathway atlas; registering patient data with the three-dimensional symptom network library to analyze how the at least one electrode maps to it; And based on the baseline symptom severity characteristics of each patient, the DBS parameters to be applied are derived and suggested.

Images