BCI Award 2022 – Check Out The Complete List Of Winners


The winners of the 2022 BCI Award were announced on September 11 to honor the best projects from the submitted applications worldwide. This year, a hundred proposals were submitted, and 12 were chosen to present their findings during the ceremony. Nuri Firat Ince presided over the jury, which also included reviewers Nicolas Opie, Cynthia Chestek, Dora Hermes, Guiseppe Pellizzer, David Moses, and Abdelkader N. Belkacem. 

The International BCI Award is among the highest honors bestowed to recognize the most innovative and outstanding research in the brain-computer interface field. A panel of leading scientists from all over the world evaluates all of the applications and selects 12 finalists. All finalists were asked to give short presentations on their projects at the awards event and to contribute a chapter to the Springer book series Brain-Computer Interface Research: A State-of-the-Art. The winners of the ceremony receive a substantial cash prize and a keynote talk invitation to the BCI & Neurotechnology Spring School 2023. The BCI award is sponsored by g.tec medical engineering, IEEE Brain, NeuroTechX, CorTec, and Intheon.

Let’s have a look at this year’s awardees.

Walking Naturally After Spinal Cord Injury Using A Brain-Spine Interface

The ceremony’s top honoree presented a brain-spine interface that stimulated the lumbar spinal cord region to activate the motor output, allowing a participant who had been paralyzed for years to walk again. The 16-electrode array was surgically placed into the spinal column’s dorsal root for stimulation. In order to help the person walk, a sequence of activations was transmitted to an array that, in turn, generates various motor outputs. These activation patterns were based on the motor intentions recorded from the electrode implanted in the motor cortex and decoded using a Markov-switching linear model. They have also combined the aforementioned components into a BCI walker, allowing the user to adjust certain parameters in order to make training and daily use easier.  

Read about the research in more detail here.

A High-Performance Intracortical Speech BCI

The second place was shared by two equally brilliant research studies. One of them is a high-performance speech BCI project that can restore rapid communication to people who are no longer able to communicate. In this project, the brain activity of an ALS patient, who was unable to speak intelligently, was recorded using microelectrode arrays. The patient was already implanted with four microelectrode arrays around the speech-related area of the cortex. The brain activity was then used to decode what the person was trying to say. The decoding model estimates threshold crossings and spike power, which are then used to train a recurrent neural network model that provides the likelihood of each phoneme at each timestep. The phonemes are then used as input into a language model, which determines the most likely word sequence the speaker intended to convey. The text that was inferred was shown to the user and also spoken out by means of a regular text-to-speech program. The model outperformed its predecessors in speech decoding by a significant margin, with an improved performance of 62 words per minute and a word error rate of 24%. Another first with this model is that, unlike past state-of-the-art models, it does not impose limitations on the vocabulary set or the sentence set.

Catch the presentation of the research study here

An Implantable Brain-Body Interface Using Intrafascicular Stimulation To Restore Hand Functions

The second position was shared by an innovative and outstanding BCI project for restoring hand movement. The goal of the study was to create a unique, accurate, and resilient motor decoding model, as well as a stimulation method that is selective and fatigue-resistant. The high-dimensional neural activity data is mapped to a low-dimensional latent dynamics or neural manifold. These intrinsic neural ensemble dynamics were directly coupled linearly to motion commands. This strategy was stable and required less calibration during the whole training and testing process. That’s a tremendous step forward because the substantial inter-session variability of brain activity has been a major roadblock in the way of bringing these devices into therapeutic use. This decoding approach was integrated with intrafascicular PNS (Peripheral Nerve Stimulation) to trigger hand movements. The advantage of using the peripheral nerve to trigger movement is that it requires a single electrode to be inserted in the nerve, and a wide range of wrist and finger movements can be produced with low amounts of current. A monkey was used to test and experiment on the entire module while doing a read-and-grasp activity. 

Learn more details about the decoding model and the stimulation approach here

Catch the video presentation of the study here.

Highly Generalizable Spelling Using A Silent-Speech BCI In A Person With Severe Anarthria

The third prize went to the neuroprosthetics research that has the potential to help a person with severe anarthria and paralysis regain the ability to speak. This project was an extension of a previous study by the same group, where they decoded the speech based on a 50-word vocabulary. In this study, they extended the vocabulary set to 1152 words that covered over 85% of the content in natural English sentences. In addition to that, this time they decoded silent-attempted speech for more clinical viability. Each letter in the English alphabet list was assigned a code word (for example, “alpha” for “A”). The participant tries to silently speak the code word while their electrocorticographic activity is being recorded. This activity was then decoded, and the letter sequences were inferred using deep learning and language models. Sentences were framed from these letter sequences based on the word vocabulary. Additionally, an attempted hand signal was employed to inform the deep-learning model that the sentence had reached its end. A 9000-word vocabulary was used to verify this spelling method in offline simulations. The median character error rate in the online simulation was 6.13%, at a pace of 29.4 characters per minute.

Read about the model and results in more detail here.

Related articles

Recent articles