In recent years, deep learning algorithms have become increasingly more prominent for their unparalleled ability to automatically learn discriminant features from large amounts of data. However, within the field of electromyographybased gesture recognition, deep learning algorithms are seldom employed as they require an unreasonable amount of effort from a single person, to generate tens of thousands of examples. This work’s hypothesis is that general, informative features can be learned from the large amounts of data generated by aggregating the signals of multiple users, thus reducing the recording burden while enhancing gesture recognition. Consequently, this paper proposes applying transfer learning on aggregated data from multiple users, while leveraging the capacity of deep learning algorithms to learn discriminant features from large datasets. Two datasets comprised of 19 and 17 able-bodied participants respectively (the first one is employed for pre-training) were recorded for this work, using the Myo Armband. A third Myo Armband dataset was taken from the NinaPro database and is comprised of 10 able-bodied participants. Three different deep learning networks employing three different modalities as input (raw EMG, Spectrograms and Continuous Wavelet Transform (CWT)) are tested on the second and third dataset. The proposed transfer learning scheme is shown to systematically and significantly enhance the performance for all three networks on the two datasets, achieving an offline accuracy of 98.31% for 7 gestures over 17 participants for the CWT-based ConvNet and 68.98% for 18 gestures over 10 participants for the raw EMG-based ConvNet. Finally, a use-case study employing eight able-bodied participants suggests that real-time feedback allows users to adapt their muscle activation strategy which reduces the degradation in accuracy normally experienced over time.

Radar-based microwave imaging has been widely studied for breast cancer detection in recent times. Sensing dielectric property differences of tissues has been studied over a wide frequency band for this application. We design single- and dual-polarization antennas for wireless ultrawideband breast cancer detection systems using an inhomogeneous multilayer model of the human breast. Antennas made from flexible materials are more easily adapted to wearable applications. Miniaturized flexible monopole and spiral antennas on a 50-μm Kapton polyimide are designed, using a high-frequency structure simulator, to be in contact with biological breast tissues. The proposed antennas are designed to operate in a frequency range of 2-4 GHz (with reflection coefficient (S11) below -10 dB). Measurements show that the flexible antennas have good impedance matching when in different positions with different curvature around the breast. Our miniaturized flexible antennas are 20 mm × 20 mm. Furthermore, two flexible conformal 4 × 4 ultrawideband antenna arrays (single and dual polarization), in a format similar to that of a bra, were developed for a radar-based breast cancer detection system. By using a reflector for the arrays, the penetration of the propagated electromagnetic waves from the antennas into the breast can be improved by factors of 3.3 and 2.6, respectively.

In this work, we present a clinical prototype with a wearable patient interface for microwave breast cancer detection. The long-term aim of the prototype is a breast health monitoring application. The system operates using multistatic time-domain pulsed radar, with 16 flexible antennas embedded into a bra. Unlike the previously reported, table-based prototype with a rigid cup-like holder, the wearable one requires no immersion medium and enables simple localization of breast surface. In comparison with the table-based prototype, the wearable one is also significantly more cost-effective and has a smaller footprint. To demonstrate the improved functionality of the wearable prototype, we here report the outcome of daily testing of the new, wearable prototype on a healthy volunteer over a 28-day period. The resulting data (both signals and reconstructed images) is compared to that obtained with our table-based prototype. We show that the use of the wearable prototype has improved the quality of collected volunteer data by every investigated measure. This work demonstrates the proof-of-concept for a wearable breast health monitoring array, which can be further optimized in the future for use with patients with various breast sizes and tissue densities.

We present a multichip structure assembled with a medical-grade stainless-steel microelectrode array intended for neural recordings from multiple channels. The design features a mixed-signal integrated circuit (IC) that handles conditioning, digitization, and time-division multiplexing of neural signals, and a digital IC that provides control, bandwidth reduction, and data communications for telemetry toward a remote host. Bandwidth reduction is achieved through action potential detection and complete capture of waveforms by means of onchip data buffering. The adopted architecture uses high parallelism and low-power building blocks for safety and long-term implantability. Both ICs are fabricated in a CMOS 0.18-mum process and are subsequently mounted on the base of the microelectrode array. The chips are stacked according to a vertical integration approach for better compactness. The presented device integrates 16 channels, and is scalable to hundreds of recording channels. Its performance was validated on a testbench with synthetic neural signals. The proposed interface presents a power consumption of 138 muW per channel, a size of 2.30 mm², and achieves a bandwidth reduction factor of up to 48 with typical recordings.

This paper presents a real-time fine gesture recognition system for multi-articulating hand prosthesis control, using an embedded convolutional neural network (CNN) to classify hand-muscle contractions sensed at the forearm. The sensor consists in a custom non-intrusive, compact, and easy-to-install 32-channel high-density surface electromyography (HDsEMG) electrode array, built on a flexible printed circuit board (PCB) to allow wrapping around the forearm. The sensor provides a low-noise digitization interface with wireless data transmission through an industrial, scientific and medical (ISM) radio link. An original frequency-time-space cross-domain preprocessing method is proposed to enhance gesture-specific data homogeneity and generate reliable muscle activation maps, leading to 98.15% accuracy when using a majority vote over 5 subsequent inferences by the proposed CNN. The obtained real-time gesture recognition, within 100 to 200 ms, and CNN properties show reliable and promising results to improve on the state-of-the-art of commercial hand prostheses. Moreover, edge computing using a specialized embedded artificial intelligence (AI) platform ensures reliable, secure and low latency real-time operation as well as quick and easy access to training, fine-tuning and calibration of the neural network. Co-design of the signal processing, AI algorithms and sensing hardware ensures a reliable and power-efficient embedded gesture recognition system.

In the realm of surface electromyography (sEMG) gesture recognition, deep learning algorithms are seldom employed. This is due in part to the large quantity of data required for them to train on. Consequently, it would be prohibitively time consuming for a single user to generate a sufficient amount of data for training such algorithms. In this paper, two datasets of 18 and 17 able-bodied participants respectively are recorded using a low-cost, low-sampling rate (200Hz), 8-channel, consumer-grade, dry electrode sEMG device named Myo armband (Thalmic Labs). A convolutional neural network (CNN) is augmented using transfer learning techniques to leverage inter-user data from the first dataset and alleviate the data generation burden imposed on a single individual. The results show that the proposed classifier is robust and precise enough to guide a 6DoF robotic arm (in conjunction with orientation data) with the same speed and precision as with a joystick. Furthermore, the proposed CNN achieves an average accuracy of 97.81% on seven hand/wrist gestures on the 17 participants of the second dataset.