Project overview - B-Cratos

B-CRATOS project overview

To better introduce our project, we decided to make a “map”.

You can click to open and close each circle to get an overview of the processes. They are described in more details in the page content.

Implant unit, electronic block diagrams, and External support

The implant electronics use a custom-designed ASIC chip to perform brain signal acquisition. Neural signals are converted to digital data for transmission and delivered to the communication unit.

The communication unit uses a high data rate backscatter link for wireless brain neural readout and another two-way low data rate for stimulation commands and implant telemetry, which also performs the WPT function. The external unit reads brain data using the RF backscattering technique, sends AI Module-generated stimulation commands, and reads implant monitoring data using another transceiver wirelessly. The neural brain data is transferred to the AI Module through the FAT-IBC channel.

High data rate neural brain data read out through backscatter channel of 32 Mbps in power consumption of several nano watts.
Ultra-low power consumption two-way wireless link up to 400 Kbps for stimulating brain by external commands, implant condition data telemetry for monitoring and facilitating wireless power transfer to enable the implant to function as a battery-free device.
Very low latency wireless stimulation link ( less than 20 msec from Mia hand sensing sensors to brain touching electrode)
High temporal and spatial resolution of brain neural data.
Wirelessly programable and configurable implant.

Fat-IBC

One of the fundamental elements forming the fat-IBC platform is the transmitting/receiving antenna, properly designed to establish an efficient communication link inside the subcutaneous fat layer. In a long-term perspective, the B-CRATOS project envisions the creation of an in-body “closed loop” between the brain and the prosthesis, using antenna elements that can be directly implanted in the fat layer. At present, non-invasive on-body antennas have been designed to demonstrate the feasibility of B-CRATOS by performing experiments with non-human primates (NHPs). The optimization process of such antennas was challenging, since these solutions are conceived to be in contact with the skin and are optimized with the aim of maximizing the radiation coupling through the subcutaneous fat layer. Different antenna solutions have been proposed and experimentally realized to test their performance.The solution reported in Figure 1 is a printed monopole antenna with a triangular-shaped radiating element and a trimmed back-placed ground plane, printed on a substrate of Rogers RT/duroid and embedded into a rigid brick made with the same dielectric material [1]. The geometrical dimensions of the monopole antenna have been optimized to maximize the transmission coefficient of a communication link made of two antennas placed on a multilayer structure mimicking the skin, fat, and muscle tissues, at the operating frequency of 2.44 GHz.

Figure 1 – Left: model of the monopole antenna; Right: visualization of the considered three-layer optimization setup.

Another implemented antenna solution is the patch antenna reported in Figure 2, where the metal patch has been optimized using a genetic algorithm interfaced with full-wave simulations performed in COMSOL Multiphysics [2]. Specifically, the metal patch has been divided into small square patches (pixels), each of them associated to a bit, with 1 and 0 indicating the presence and the absence of metal, respectively. Each sequence of bits corresponds to a combination of pixels, and hence to a particular layout for the antenna solution. The pixelated patch has been optimized to maximize the transmission coefficient within a sufficiently large bandwidth for a link of two antennas placed on a three-layer phantom.

Figure 2 – Left: model of the pixelated patch antenna solution; Right: optimization setup.

Both the antenna solutions presented so far have been experimentally realized by Fondazione LINKS, Turin, and tested at the University of Uppsala on a properly developed three-layer phantom model emulating the dielectric properties of skin, fat, and muscle [3][4][5][6][7], inserted in a shielded portable chamber. The results will be presented at the 18th European Conference on Antennas and Propagation (EuCAP), in Glasgow, on 17-22 March 2024 [8].

[1] R. Gaffoglio, G. Giordanengo, L. Teodorani, M. Righero, G. Musacchio Adorisio, B. Mandal, M. D. Perez, R. Augustine, and G. Vecchi, “Compact optimized antenna solution for radiation coupling improvement in the subcutaneous fat layer”, 17th European Conference on Antennas and Propagation (EuCAP), 2023.

[2] R. Gaffoglio, G. Giordanengo, M. Righero, G. Musacchio Adorisio, B. Mandal, M. D. Perez, H. Scherberger, R. Augustine, and G. Vecchi, “Pixel Optimization via Genetic Algorithm of a Flat Epidermal Antenna for Fat Channel Communication”, TechRxiv. February 07, 2023.

[3] P. K. B. Rangaiah, J. Engstrand, T. Johansson, M. D. Perez, and R. Augustine, “92 Mb/s fat-intrabody communication (fat-IBC) with low-cost WLAN hardware,” IEEE. Trans. Biomed. Eng., vol. 71, no. 1, 2024.

[4] L. Joseph, M. D. Perez, and R. Augustine, “Development of ultrawideband 500 MHz-20 GHz human skin phantoms for various microwaves based biomedical applications,” IEEE Conference on Antenna Measurements & Applications (CAMA), pp. 1-3, 2018.

[5] Asan, N.B., Noreland, D., Hassan, E., Redzwan Mohd Shah, S., Rydberg, A., Blokhuis, T.J., Carlsson, P.-O., Voigt, T. and Augustine, R. (2017), Intra-body microwave communication through adipose tissue. Healthcare Technology Letters, 4: 115-121. https://doi.org/10.1049/htl.2016.0104

[6] Asan, N.B.; Hassan, E.; Velander, J.; Mohd Shah, S.R.; Noreland, D.; Blokhuis, T.J.; Wadbro, E.; Berggren, M.; Voigt, T.; Augustine, R. Characterization of the Fat Channel for Intra-Body Communication at R-Band Frequencies. Sensors 2018, 18, 2752. https://doi.org/10.3390/s18092752

[7] N. B. Asan et al., “Data Packet Transmission Through Fat Tissue for Wireless IntraBody Networks,” in IEEE Journal of Electromagnetics, RF and Microwaves in Medicine and Biology, vol. 1, no. 2, pp. 43-51, Dec. 2017, doi: 10.1109/JERM.2017.2766561

[8] R. Gaffoglio, G. Giordanengo, G. Musacchio Adorisio, B. Mandal, J. Engstrand, R. Augustine, and G. Vecchi, “Compact Antenna Solutions for Data Transmission Using Fat-Intrabody Communication (Fat-IBC)”, 18th European Conference on Antennas and Propagation (EuCAP), 2024 (accepted)

AI-Module

AI – Module

Accurate control of a robotic prosthesis requires reliable decoding of neural data, which provides a mapping between the signals acquired by the brain implant and the prosthesis control variables. While the state of practice in neural decoding involves well-known approaches using linear models and Bayesian classifiers [1,2,3], these models are hardly capable of capturing the complexity of the information encoded in neural signals and typically require the patient to learn to drive the prosthesis. The goal of this work is to leverage the recent advancements in deep learning to implement accurate control based on modern techniques like recurrent neural networks and long short-term memory (LSTM) networks to enable a more natural and robust control of the device.

The developed decoding algorithm leverages a bidirectional LSTM neural network architecture and is trained on High-Performance Computing facilities and deployed on a dedicated board that consumes less than 5W.

Currently, we successfully demonstrated that LSTM models can outperform classical approaches for pre-recorded neural data [4].

At the same time, the AI module plays a key role in conveying tactile feedback from the E-skin to the brain. To achieve this, the AI module board is also connected to the sensors of the sensorized skin via digital channels providing contact/pressure information encoded in spike rates between 0 and 200Hz. These signals are acquired by the board, converted into pre-defined stimulation patterns, and relayed to the implant via the FAT-IBC.

[1] J. I. Glaser, A. S. Benjamin, R. H. Chowdhury, M. G. Perich, L. E. Miller, and K. P. Kording, “Machine Learning for Neural Decoding,” eNeuro, vol. 7, no. 4, p. ENEURO.0506-19.2020, Jul. 2020, doi: 10.1523/ENEURO.0506-19.2020.

[2] J. A. Livezey and J. I. Glaser, “Deep learning approaches for neural decoding across architectures and recording modalities,” Briefings in Bioinformatics, vol. 22, no. 2, pp. 1577–1591, Mar. 2021, doi: 10.1093/bib/bbaa355.

[3] S. Schaffelhofer, A. Agudelo-Toro, and H. Scherberger, “Decoding a Wide Range of Hand Configurations from Macaque Motor, Premotor, and Parietal Cortices,” Journal of Neuroscience, vol. 35, no. 3, pp. 1068–1081, Jan. 2015, doi: 10.1523/JNEUROSCI.3594-14.2015.

[4] P. Viviani et al., “Deep Learning for Real-Time Neural Decoding of Grasp,” in Machine Learning and Knowledge Discovery in Databases: Applied Data Science and Demo Track, vol. 14174, G. De Francisci Morales, C. Perlich, N. Ruchansky, N. Kourtellis, E. Baralis, and F. Bonchi, Eds., in Lecture Notes in Computer Science, vol. 14174. , Cham: Springer Nature Switzerland, 2023, pp. 379–393. doi: 10.1007/978-3-031-43427-3_23.

Prosthetic hand with self powered electronic skin

Modern robotic prostheses still suffer from many limitations, among them due to the lack of sensory feedback, poor functionality, and cosmesis, as well as wearing discomfort. Recently, the development of robotic prostheses has mostly focused on increasing the number of degrees of freedom, trying to satisfy appropriate size and weight requirements. However, a recent study highlights how off-the-shelf multi-grip hand prostheses did not show relevant differences in outcomes compared to a standard hand prosthesis (which provides a single primitive grasping action) if controlled by means of standard human machine interface.

B-CRATOS aims at developing machine learning/AI techniques for real-time decoding of neural signals into hand actions – see the part about AI-Module.

The hand prosthesis will be endowed with an artificial skin with self-powered tactile sensors., The skin is designed to efficiently process and delivery dynamic tactile information about touch and grasp. An attempt of integration the skin in the thumb, index and medial fingers of the MIA hand prosthesis will be made to feedback meaningful information to the brain cortex.

Non-Human-Primate training and testing

iBCIs

Picture a world where people with paralysis or other disabilities can regain control of their limbs or interact with the world around them simply by thinking. This vision is becoming increasingly realistic thanks to the advancements in brain-computer interfaces (BCIs) and specifically, intracortical BCIs (iBCIs). iBCIS directly measures brain activity by implanting tiny electrodes into the brain.

Animal models

To demonstrate and advance iBCIs, animal models are required, preventing risks to human users. Particularly, non-human primates (NHPs) like macaques are part of these experiments. NHPs share many similarities with humans in terms of brain structure and function, making them valuable models for BCI research. Studies in NHPs have helped decipher the neural signals associated with movement, intention, and other cognitive processes, providing invaluable insights for developing effective BCI systems.

Real-time

Beyond animal modes, real-time, online experimental demonstrations play a crucial role in addressing the challenges in iBCIs development. By enabling researchers to observe and analyze brain activity in real-time, these demonstrations allow for rapid iteration and refinement of BCI algorithms. This feedback loop is essential for developing BCIs that can reliably decode complex intentions and control prosthetic devices with precision.

The role of DPZ

For the envisioned clinical application in BCRATOS, three elements are required:

High bandwidth (# of channels) neural data (BRME, NTNU, SSS, LINKS)
Direct somatosensory feedback (tactile, proprioceptive) (BRME, SSA, UU) are required in order to achieve the higher degrees of freedom required for hand grasping tasks.
High-bandwidth wireless communication (NTNU, UU, LINKS) is also required for safety (lower infection risk) and usability.

DPZ will demonstrate all three of these requirements using the B-CRATOS system first in NHP cadavers and then in awake behaving NHPs trained in a task using arm control signals from the motor cortex (M1) and tactile feedback via stimulation in sensory cortex (S1). Individual components of the system will be first demonstrated using a percutaneous connector before demonstrating the final integrated implant.

Experimental setup

To demonstrate real-time control with monkeys, we require an experimental setup that combines the B-CRATOS system with our system and complies with strict timing restrictions. Additionally, as part of the task, we are responsible for a robot arm that can position the MIA hand. For these we are programming a Franka-Emika Panda Research arm. The system design is being tested and integrated is presented in the following image.