Utility of Imaging Predictors in Deep Brain Stimulation

10 

Abstract Submission 

Patricia Zvarova¹, Ningfei Li¹, Ilkem Sahin¹, Barbara Hollunder¹, Martin Reich², Jens Volkmann², Vincent Odekeren³, Rob de Bie³, Xin Xu⁴, Zhipei Ling⁵, Chen Yao⁶, Andrea Kühn¹, Nanditha Rajamani¹, Andreas Horn⁷

¹Department of Neurology, Charité – Universitätsmedizin Berlin, Berlin, Germany, ²Department of Neurology, Universitätsklinikum Würzburg, Würzburg, Germany, ³Department of Neurology, Academisch Medisch Centrum Universiteit van Amsterdam, Amsterdam, Netherlands, ⁴Department of Neurosurgery, Chinese PLA General Hospital, Beijing, Beijing, China, ⁵Department of Neurosurgery, Hainan Hospital of Chinese PLA General Hospital, Hainan, China, ⁶Department of Neurosurgery, Shenzhen Key Laboratory of Neurosurgery, Shenzhen, China, ⁷Brigham and Women’s Hospital, Harvard Medical School, Boston, MA

Patricia Zvarova  
Department of Neurology, Charité – Universitätsmedizin Berlin
Berlin, Germany

Ningfei Li  
Department of Neurology, Charité – Universitätsmedizin Berlin
Berlin, Germany

Ilkem Sahin  
Department of Neurology, Charité – Universitätsmedizin Berlin
Berlin, Germany

Barbara Hollunder  
Department of Neurology, Charité – Universitätsmedizin Berlin
Berlin, Germany

Martin Reich  
Department of Neurology, Universitätsklinikum Würzburg
Würzburg, Germany

Jens Volkmann  
Department of Neurology, Universitätsklinikum Würzburg
Würzburg, Germany

Vincent Odekeren  
Department of Neurology, Academisch Medisch Centrum Universiteit van Amsterdam
Amsterdam, Netherlands

Rob de Bie  
Department of Neurology, Academisch Medisch Centrum Universiteit van Amsterdam
Amsterdam, Netherlands

Xin Xu  
Department of Neurosurgery, Chinese PLA General Hospital, Beijing
Beijing, China

Zhipei Ling  
Department of Neurosurgery, Hainan Hospital of Chinese PLA General Hospital
Hainan, China

Chen Yao  
Department of Neurosurgery, Shenzhen Key Laboratory of Neurosurgery
Shenzhen, China

Andrea Kühn  
Department of Neurology, Charité – Universitätsmedizin Berlin
Berlin, Germany

Nanditha Rajamani  
Department of Neurology, Charité – Universitätsmedizin Berlin
Berlin, Germany

Andreas Horn  
Brigham and Women’s Hospital, Harvard Medical School
Boston, MA

The clinical efficacy of deep brain stimulation (DBS) is significantly influenced by the accuracy of electrode placement (Okun et al., 2005, 2008; Bot et al., 2018; Neudorfer et al., 2023). Computerized models that reconstruct the electrode in silico have been used to investigate the relationship between electrode placements and clinical improvements. These models may be studied on a local (coordinates and stimulation volumes) and global network level (activated fiber tracts and associated functional networks). If accurate, models may be used to predict clinical outcomes in patients that were not used to create them (Horn et al., 2017; Boutet et al., 2021; Roediger et al., 2022). If successful, this leads to clinical utility of the models to potentially guide both DBS surgery and programming. Given promise, a recent trend in the emergence of computerized DBS models could be seen in the literature. However, a guide for best practices and parameter choices is lacking. Here, our aim was to compare a multitude of models and parameter choices based on a large multi-centre DBS cohort of patients who underwent the subthalamic nucleus DBS (STN-DBS) surgery for Parkinson's disease (PD).

We base analyses on retrospective data from a large cohort of N = 170 PD patients who underwent bilateral DBS targeting to the STN at five international centres. We used Lead-DBS v3 to localize electrodes and model stimulation volumes in each patient (Neudorfer et al., 2023). Relative motor improvements before and after surgery were measured via the Unified Parkinson's Disease Scale III (UPDRS-III). We then created predictive models on four levels: The first model was solely based on active coordinates of stimulation sites. Euclidean distance of active electrode contacts from an a priori optimal target coordinate (Caire et al., 2013) was calculated. The second model included an estimated stimulation volume based on programming parameters and was based on overlaps between these volumes and a sweet spot that was calculated following the approach by Horn et al., (2022). Third and fourth, DBS Fiber Filtering (Irmen et al., 2020) and DBS Network Mapping (Horn et al., 2017) approaches were carried out to identify the streamlines and functional networks associated with beneficial clinical effects (Figure 1). Sweet spots, optimal streamlines and optimal networks were validated using leave-nothing-out and 10-fold cross-validation to avoid circularity of the models. Results from all four models were fed into a multiple linear regression analysis.

As expected, Euclidean distances between active contacts and the a priori optimal coordinate correlated negatively with clinical improvements (R = -0.30, p < 0.001). Similarly, sweet spot, optimal tract and optimal network models were able to explain significant amounts of variance in clinical improvements (R = 0.19, p = 0.015; R = 0.29, p < 0.001 and R = 0.34, p < 0.001, respectively). In a combined model, 14% of the variance in empirical clinical improvements could be explained (F-statistic = 6.8, p < 0.001). In this model, only the Fiber Filtering scores contributed significantly (ß = 0.21, p = 0.018) (Figure 2).

Our findings confirm a robust relationship between electrode placements and clinical improvements following STN-DBS for PD. When combining results of models from local and global measures, models can still only account for a fraction of the overall variance.

Deep Brain Stimulation ¹

Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s)

Connectivity (eg. functional, effective, structural)

fMRI Connectivity and Network Modeling

Other Methods ²

Data analysis

Degenerative Disease

Movement Disorder

MRI

Open-Source Software

Statistical Methods

Other - DBS