Brain computer interfaces (BCIs) are devices capable of relaying information directly from the brain to a digital device. BCIs have been proposed for a diverse range of clinical and commercial applications; for example, to allow paralyzed subjects to communicate, or to improve machine human interactions. At their core, BCIs need to predict the current state of the brain from variables measuring functional physiology. Functional near infrared spectroscopy (fNIRS) is a non-invasive optical technology able to measure hemodynamic changes in the brain. Along with electroencephalography (EEG), fNIRS is the only technique that allows non-invasive and portable sensing of brain signals. Portability and wearability are very desirable characteristics for BCIs, as they allow them to be used in contexts beyond the laboratory, extending their usability for clinical and commercial applications, as well as for ecologically valid research. Unfortunately, due to limited access to the brain, non-invasive BCIs tend to suffer from low accuracy in their estimation of the brain state. It has been suggested that feedback could increase BCI accuracy as the brain normally relies on sensory feedback to adjust its strategies. Despite this, presenting relevant and accurate feedback in a timely manner can be challenging when processing fNIRS signals, as they tend to be contaminated by physiological and motion artifacts.
In this dissertation, I present the hardware and software solutions we proposed and developed to deal with these challenges. First, I will talk about ninjaNIRS, the wearable open source fNIRS device we developed in our laboratory, which could help fNIRS neuroscience and BCIs to become more accessible. Next, I will present an adaptive filter strategy to recover the neural responses from fNIRS signals in real-time, which could be used for feedback and classification in a BCI paradigm.
We showed that our wearable fNIRS device can operate autonomously for up to three hours and can be easily carried in a backpack, while offering noise equivalent power comparable to commercial devices. Our adaptive multimodal Kalman filter strategy provided a six-fold increase in contrast to noise ratio of the brain signals compared to standard filtering while being able to process at least 24 channels at 400 samples per second using a standard computer. This filtering strategy, along with visual feedback during a left vs right motion imagery task, showed a relative increase of accuracy of 37.5% compared to not using feedback. With this, we show that it is possible to present relevant feedback for fNIRS BCI in real-time. The findings on this dissertation might help improve the design of future fNIRS BCIs, and thus increase the usability and reliability of this technology.
| Identifer | oai:union.ndltd.org:bu.edu/oai:open.bu.edu:2144/45483 |
| Date | 17 January 2023 |
| Creators | Ortega, Antonio |
| Contributors | Boas, David A. |
| Source Sets | Boston University |
| Language | en_US |
| Detected Language | English |
| Type | Thesis/Dissertation |
Page generated in 0.0017 seconds