Inspired by @andreashorn.org, the core is connectomics-informed decoding: each electrode contact is mapped to a “fingerprint” of whole-brain connectivity, which predicts decoding performance. This enables out-of-cohort generalization without patient-specific training. is the innovator.

For epilepsy, we optimized seizure detection using AI-guided parameter discovery. Connectomics revealed a distributed seizure network (hippocampus, cingulate, occipital cortex), enabling more accurate, personalized detection. Our parameters were constrained to device capabilities.

In depression, we decoded emotional states from the subgenual cingulate. Better decoding correlated with clinical outcomes after DBS, and network maps aligned with known TMS targets. A link between invasive decoding and non-invasive therapy for the next-generation of AI inspired neuromodulation.

For movement disorders, connectomics-informed decoders generalized across Parkinson’s and epilepsy patients. Plug-and-play decoding without per-patient calibration triggered adaptive DBS in a newly recruited patient. #cebra outperformed other decoder architectures.

Traditional brain–computer interfaces (BCIs) often need individual retraining for every patient. Our approach shows that generalizable decoders can work across cohorts in Europe, the US, and China—sometimes even outperforming individual models. The platform translates offline data to real-time.

Super honored to be invited by Stephanie Lacour to speak at
@neuroxepfl.bsky.social on "Deep Brain Stimulation as a Therapeutic Brain-Computer Interface: From Disease Signatures to AI-Driven Neural Circuit Prosthetics" this Thursday. Open on Zoom, too! Info here: memento.epfl.ch/event/neuro-...

Super cool paper on the utility of Reinforcement learning models for understanding dopaminergic learning in perceptual decision making.

The paper provides the conceptual foundation for @ERC_Research starting grant #ReinforceBG which started in March 2023 in the @ICNeuromodulate at @ChariteBerlin for which Alessia Cavallo joined as a @bccn_berlin MSc. now pursuing her PhD. 6/6

Finally, we provide an outlook on how we may use deep brain stimulation as a #neuroprosthetics approach to mimic dopamine signaling with temporally precise #neurostimulation power by #AI and #machinelearning to counteract #PD symptoms with invasive #neurotechnology. 5/n

Use it to lose it: We argue that neural reinforcement may be a key to understanding #Parkinsons. Are PD motor signs a consequence of previous movement in absence of dopamine, as suggested in a recent rodent paper ? doi.org/10.1073/pnas...
We believe this is worth investigating in humans! 4/n

We take up aspects of the concept of neural #reinforcement recently highlighted by @vr_athalye, @blancinegre1972 & @ruimcosta here doi.org/10.1016/j.co... - reiterating the role of #dopamine for invigoriation and consolidation of neural circuit dynamics. 3/n

In our paper, we explore dopamine's multifaceted influence on the basal ganglia, impacting everything from #attention and #sleep to #motivation and #learning. We discuss overarching circuit computations underlying these functions which may explain #Parkinsons symptoms. 2/n

Thrilled to be invited to the Trailblazer in Neuroscience series by European Journal of Neuroscience! PhD student Alessia Cavallo and I share our view on basal ganglia reinforcement for Parkinsons disease with opportunities arising for neurotechnology. doi.org/10.1111/ejn....

1/n

Finally, we benchmark seizure decoding for embedded devices in epilepsy patients with Neuropace RNS brain implants. We show that advanced machine learning parametrization that is implementable for validation in the embedded device itself, can improve decoder accuracy.

Next, we decoded emotions in patients undergoing subgenual cingulate DBS for depression. Decoder accuracy correlated with symptom alleviation. With @andreashorn.bsky.social Lead-DBS connectomics toolbox, we identified a left-lateralized prefrontal circuit encoding emotional valence in invasive data.

Our results show that the future of BCI will not require tedious patient individual training, but also warn us that severity of brain disorders and electrical stimulation artifacts can hamper decoder performance, e.g. through a negative correlation of PD symptoms with accuracy.

First, we developed and prospectively validated plug & play movement decoders combining connectomics with contrastive learning in CEBRA for ECoG signals. This  image shows decoding in a prospective validation of a model that has never seen brain data from the patient before.

We describe a brain computer interface strategy that enables utilization of large datasets facilitating machine learning on divers populations across continents (USA, Europe, China), brain disorders (Parkinson, depression, epilepsy) and tasks or symptoms (movement, emotion, seizures).

Preprint 🚨 Invasive #neurotechnology has potential for breakthroughs in the treatment of brain disorders. To unlock this potential we developed a platform that combines MRI connectomics with neurophysiology for circuit discovery and decoding of movement, emotion and epilepsy! bit.ly/invasiveBCI