Deep Brain Stimulation (DBS) has shown particular effectiveness in treating symptoms associated with movement disorders, but uncertainty still looms over the most effective spots to target, and surgeons need a more automated way of testing the implanted electrodes. As part of his AFR PhD project at the Centre Hospitalier de Luxembourg (CHL), computer scientist Dr Andreas Husch developed innovative image-based computational approaches to aid DBS, which are already being used by scientists across the world.
Deep brain stimulation (DBS) is an intervention based on implanting permanent electrodes into the human brain to alleviate the symptoms of various movement and neuropsychiatric disorders, such as tremors, dystonia and Parkinson’s disease.
DBS has three phases, with the actual programming of the electrode taking place at the post-operative stage:
“A deep brain stimulator is a bit like a pacemaker for the heart: it’s normally placed in the chest area cable running to the brain under the skin. The electrode in the brain has a pole in it, like a hose. After the surgery, the pole is removed and the electrode is left fully flexible”, Dr Andreas Husch explains, adding:
“When you implant electrodes in the brain, air gets around the brain, which moves its position. After one or two weeks, the brain is back in its original position, pulling the electrodes with it, almost bending them a bit, hence the need for them to be flexible. The electrode has multiple contacts, which are connected to the ‘pacemaker’. The pacemaker has to be programmed in order to define which of the four contacts is stimulated, and how strongly. The stimulation can be adapted to the patient to get the best effect and ideally no side effects.”
The first DBS experiments took place about 60 years ago, modern DBS treatments have been around for some 30 years – but open questions remain. A key problem in DBS research is the question of the precise definition of the optimal stimulation targets for the treatment of different diseases, particularly for Parkinson’s disease.
From methods for flood prediction to computer aided deep brain stimulation
Dr Andreas Husch is a computer scientist – his Bachelor focused on Software Engineering, while his Masters focused on Machine Learning and Imaging. Already during his Bachelors (B.Sc.) at the University of Applied Sciences in Trier, Andreas started working as a student research assistant, on a project looking at flood prediction for the Moselle River. The link between flood prediction and DBS is not obvious, but Andreas explains a common thread:
“When you abstract enough in computer science, you often find that things share a common structure. For example, in a river, you can measure the water gauge changing over time. In the brain, you can measure microelectric recordings from single neuron cells, which is a signal over time – so data-wise, it’s very similar. This means you can apply similar algorithms.”
During his Masters’, Andreas worked as a research assistant with Computer Science Professor Peter Gemmar, who was involved in projects relating to DBS and flood prediction. Gemmar applied some algorithmic ideas from flood prediction to DBS in collaborative work with Prof Dr Frank Hertel, who heads up the Neurosurgical National Department of Luxemburg, at the Centre Hospitalier de Luxembourg (CHL) and expressed an urgent need to get computer scientists in a DBS operating room. Together, they successfully applied for an AFR grant, placing the computer scientist in a clinical setting:
“My project was highly translational – I spent a lot of time observing surgeries, trying to understand the needs of the surgeons – essentially requirements engineering: what do they really need and how can we help with computer science,” Andreas explains.
Responding to a need for post-operative aids
“Originally, my project was meant to focus on improving the intra-operative phase, but during the project new developments in the field changed the focus to post-operative,” Andreas explains. The reason was new developments with electrodes for DBS use: Traditionally, four contacts are implanted in the brain for DBS, however, now it is possible to have as many as eight with more complex shares, enabling a “steering” of the current in a certain direction .
“This makes the post-operative part much more complex, naturally, as you are working with more contacts and there is a significantly larger amount time required for programming the stimulator, due to a situation known as a “combinatorial explosion”. This created an urgent need to come up with post-operative aid for surgeons,” Andreas explains.
This led Andreas to come up with the ‘PaCER’ algorithm, which is currently the most precise and only fully automatic algorithm published to recover post-operative electrode from CT data. The overall goal of the project was to create the means to allow objective and as far as possible automatic assessment and support of a personalised deep brain stimulation process. These novel means contribute to a better understanding of the action of DBS in an individual subject and may enable improved therapy by providing personalised computational guidance of the physician.
To patent or open source?
Andreas explains the team had a long discussion about whether they should apply for a patent or publish PaCER as open source software, making it freely available to the scientific community. They decided to do the latter:
“We decided to open source it, we also thought this would be better for Luxembourg and could help establish us more in the scientific community”, Andreas explains.
For example, PaCER has been adopted by the LeadDBS group (coordinated at Charité Berlin), and was integrated in their leading deep brain stimulation research software package, making it accessible to researchers around the globe.
Automated solution for surgeons
Andreas explains that it takes quite some time to learn how to operate the LeadDBS tool, and that he therefore wanted something more automated to help to DBS surgeons.
“We teamed up with Mikkel Petersen from the University of Aarhus, Denmark, and tried to bring PaCER into an automatic pipeline. To help a doctor you need an estimate of the individual brain structure of the patient and bring it together with lead reconstruction.
“For this method, you combine ATLAS MRI data with PaCER data and it outputs it as a fully interactive 3D PDF. This helps you see how well placed the contacts / electrodes are. It helps give the neurologist an idea of how they can improve without having to test everything. It can also calculate some metrics, e.g. whether the contact is inside or outside of the target structure.
While this is still a prototype the idea is to show how it could be used clinically,” Andreas explains, adding that the collaborators in Denmark also ran a small independent clinical validation study, in which they were quickly able to apply the methods to their clinical data.
“This approach was published as an idea and the Danish colleagues proofed the principle clinical applicability. We have just gotten ethics approval for a larger study validating this in Luxembourg. Furthermore we are considering setting up a kind of web service. People could upload their anonymized data, it would then process for 1 – 2 hours – PACER itself is quick, but the ATLAS estimate of the brain takes quite some time – and then they get this interactive 3D PDF. We are currently teaming up with colleagues from LCSB Bioinformatics Core group to realise such a service utilising their ARTENOLIS continuous integration platform in line with their efforts for Responsible and Reproducible Research (R3).”
When asked whether PaCER could become the standard for post-operative DBS, Andreas explains that it could, but that it would still need some extensions, and company involvement to get a CE mark.
Findings awarded for improvements for post-operative DBS phase & Shanghai talk
In September 2018, Andreas Husch was presented with an award sponsored by InSilicoTrials in collaboration with the VPH Institute for his PhD work, in particular the improvements to the post-operative phase of DBS. The award is presented to individuals for outstanding achievements during their PhD thesis, specifically focusing on the translational aspects of their work.
Dr Husch was also recently invited to give a talk in Shanghai, which covered the key contributions of thesis. The talk was delivered together with Prof Dr Frank Hertel, Dr Husch’s PhD co-supervisor.
Dr Husch’s AFR PhD project was carried out at the Centre Hospitalier de Luxembourg (CHL), in collaboration with the University of Luxembourg, the LCSB and Hochschule Trier. Dr Husch is currently Postdoc at the LCSB at the University of Luxembourg.
Andreas Husch on fully automated DBS electrode reconstruction using the great PaCER method (Husch er al. 2017 NI: Clinical, Code available on github and within Lead-DBS) pic.twitter.com/vR52FW1H0x— Lead DBS (@leaddbs) September 19, 2018
PaCER algorithm by Andreas Husch (Husch et al. NI: clinical 2017) integrated into Lead-DBS 2.0 pic.twitter.com/LrhsrFQpKw— Lead DBS (@leaddbs) September 20, 2018