Search

This paper presents a reconfigurable near-sensor anomaly detection processor to real-time monitor the potential anomalous behaviors of amputees with limb prostheses. The processor is low-power, low-latency, and suitable for equipment on the prostheses and comprises a reconfigurable Variational Autoencoder (VAE), a scalable Self-Organizing Map (SOM) Array, and a window-size-adjustable Markov Chain, which can implement an integrated miniaturized anomaly detection system. With the reconfigurable VAE, the proposed processor can support up to 64 sensor sampling channels programmable by global configuration, which can meet the anomaly detection requirements in different scenarios. A scalable SOM array allows for the selection of different sizes based on the complexity of the data. Unlike traditional time accumulation-based anomaly detection methods, the Markov Chain is utilized to detect time-series-based anomalous data. The processor is designed and fabricated in a UMC 40-nm LP technology with a core area of 1.49 mm 2 and a power consumption of 1.81 mW. It achieves real-time detection performance with 0.933 average F1 Score for the FSP dataset within 24.22 μs, and 0.956 average F1 Score for the SFDLA-12 dataset within 30.48 μs. The energy dissipation of detection for each input feature is 43.84 nJ with the FSP dataset, and 55.17 nJ with the SFDLA-12 dataset. Compared with ARM Cortex-M4 and ARM Cortex-M33 microcontrollers, the processor achieves energy and area efficiency improvements ranging from 257 ×, 193 × and 11 ×, 8 ×, respectively.

Dopaminergic treatment has proved effective to Parkinson's disease (PD), but the conventional treatment assessment is human-administered and prone to intra- and inter-assessor variability. In this paper, we propose a knowledge-driven framework and discover that brain ACtivation-Transition-Spectrum (ACTS) features achieve effective quantified assessments of dopaminergic therapy in PD. Firstly, brain activities of fifty-one PD patients during clinical walking tests under the OFF and ON states (without and with dopaminergic medication) were measured with functional near-infrared spectroscopy (fNIRS). Then, brain ACTS features were formulated based on the medication-induced variations in temporal features of brain regional activation, transition features of brain hemodynamic states, and graph spectrum of brain functional connectivity. Afterwards, a prior selection algorithm was constructed based on recursive feature elimination and graph spectrum analysis for the selection of principal discriminative features. Further, linear discriminant analysis was conducted to predict the treatment-induced improvements. The results demonstrated that the proposed method decreased the misclassification probability from 42% to 16% in the evaluations of dopaminergic treatment and outperformed existing fNIRS-based methods. Therefore, the proposed method promises a quantified and objective approach for dopaminergic treatment assessment, and our brain ACTS features may serve as promising functional biomarkers for treatment evaluation.

Hand motor impairment has seriously affected the daily life of the elderly. We developed an electromyography (EMG) exosuit system with bidirectional hand support for bilateral coordination assistance based on a dynamic gesture recognition model using graph convolutional network (GCN) and long short-term memory network (LSTM). The system included a hardware subsystem and a software subsystem. The hardware subsystem included an exosuit jacket, a backpack module, an EMG recognition module, and a bidirectional support glove. The software subsystem based on the dynamic gesture recognition model was designed to identify dynamic and static gestures by extracting the spatio-temporal features of the patient's EMG signals and to control glove movement. The offline training experiment built the gesture recognition models for each subject and evaluated the feasibility of the recognition model; the online control experiments verified the effectiveness of the exosuit system. The experimental results showed that the proposed model achieve a gesture recognition rate of 96.42% ± 3.26 %, which is higher than the other three traditional recognition models. All subjects successfully completed two daily tasks within a short time and the success rate of bilateral coordination assistance are 88.75% and 86.88%. The exosuit system can effectively help patients by bidirectional hand support strategy for bilateral coordination assistance in daily tasks, and the proposed method can be applied to various limb assistance scenarios.

In most real world rehabilitation training, patients are trained to regain motion capabilities with the aid of functional/epidural electrical stimulation (FES/EES), under the support of gravity-assist systems to prevent falls. However, the lack of motion analysis dataset designed specifically for rehabilitation-related applications largely limits the conduct of pilot research. We provide an open access dataset, consisting of multimodal data collected via 16 electromyography (EMG) sensors, 6 inertial measurement unit (IMU) sensors, and 230 insole pressure sensors (IPS) per foot, together with a 26-sensor motion capture system, under different MOVEments and POstures for Rehabilitation Training (MovePort). Data were collected under diverse experimental paradigms. Twenty four participants first imitated multiple normal and abnormal body postures including (1) normal standing still, (2) leaning forward, (3) leaning back, and (4) half-squat, which in practical applications, can be detected as feedback to tune the parameters of FES/EES and gravity-assist systems to keep patients in a target body posture. Data under imitated abnormal gaits, e.g., (1) with legs raised higher under excessive electrical stimulation, and (2) with dragging legs under insufficient stimulation, were also collected. Data under normal gaits with low, medium and high speeds are also included. Pathological gait data from a subject with spastic paraplegia further increases the clinical value of our dataset. We also provide source codes to perform both intra- and inter-participant motion analyses of our dataset. We expect our dataset can provide a unique platform to promote collaboration among neurorehabilitation engineers.

Wearable intelligent health monitoring devices with on-device biomedical AI processor can be used to detect the abnormity in users' biomedical signals (e.g., ECG arrythmia classification, EEG-based seizure detection). This requires ultra-low power and reconfigurable biomedical AI processor to support battery-supplied wearable devices and versatile intelligent health monitoring applications while achieving high classification accuracy. However, existing designs have issues in meeting one or more of the above requirements. In this work, a reconfigurable biomedical AI processor (named BioAIP) is proposed, mainly featuring: 1) a reconfigurable biomedical AI processing architecture to support versatile biomedical AI processing. 2) an event-driven biomedical AI processing architecture with approximate data compression to reduce the power consumption. 3) an AI-based adaptive-learning architecture to address patient-to-patient variation and improve the classification accuracy. The design has been implemented and fabricated using a 65nm CMOS process technology. It has been demonstrated with three typical biomedical AI applications, including ECG arrythmia classification, EEG-based seizure detection and EMG-based hand gesture recognition. Compared with the state-of-the-art designs optimized for single biomedical AI tasks, the BioAIP achieves the lowest energy per classification among the designs with similar accuracy, while supporting various biomedical AI tasks.

Parkinson's disease (PD) is a prevalent brain disorder, and PD diagnosis is crucial for treatment. Existing methods for PD diagnosis are mainly focused on behavior analysis, while the functional neurodegeneration of PD has not been well investigated. This paper proposes a method to signify functional neurodegeneration of PD with dynamic functional connectivity analysis. A functional near-infrared spectroscopy (fNIRS)-based experimental paradigm was designed to capture brain activation from 50 PD patients and 41 age-matched healthy controls in clinical walking tests. Dynamic functional connectivity was constructed with sliding-window correlation analysis, and k-means clustering was applied to generate the key brain connectivity states. Dynamic state features including state occurrence probability, state transition percentage and state statistical features were extracted to quantify the variations of brain functional networks. A support vector machine was trained to classify PD patients and healthy controls. Statistical analysis was conducted to investigate the difference between PD patients and healthy controls as well as the relationship between dynamic state features and the MDS-UPDRS sub-score of gait. The results showed that PD patients had a higher probability of transiting to brain connectivity states with high levels of information transmission compared with healthy controls. The MDS-UPDRS sub-score of gait and the dynamics state features showed a significant correlation. Moreover, the proposed method had better classification performances than the available fNIRS-based methods in terms of accuracy and F1 score. Thus, the proposed method well signified functional neurodegeneration of PD, and the dynamic state features may serve as promising functional biomarkers for PD diagnosis.

Congenital Muscular Torticollis (CMT) is a neuromuscular disease in children, which leads to exacerbation of postural deformity and neck muscle dysfunction with age. Towards facilitating functional assessment of neuromuscular disease in children, topographic electromyography (EMG) maps enabled by flexible and stretchable surface EMG (sEMG) electrode arrays are used to evaluate the neck myoelectric activities in this study. Customed flexible and stretchable sEMG electrode arrays with 84 electrodes were utilized to record sEMG in all subjects during neck motion tasks. Clinical parameter assessments including the cervical range of motion (ROM), sonograms of the sternocleidomastoid (SCM), and corresponding histological analysis were also performed to evaluate the CMT. The muscle activation patterns of neck myoelectric activities between the CMT patients and the healthy subjects were asymmetric during different neck motion tasks. The CMT patients presented significantly lower values in spatial features of two-dimensional (2D) correlation coefficient, left/right energy ratio, and left/right energy difference (p < 0.001). The 2D correlation coefficient of activation patterns of neck rotation and extension in CMT patients significantly correlated with clinical parameter assessments (p < 0.05). The findings suggest that the spatial features of muscle activation patterns based on the sEMG electrode arrays can be utilized to evaluate the CMT. The flexible and stretchable sEMG electrode array is promising to facilitate the functional evaluation and treatment strategies for children with neuromuscular disease.

Brain-computer interfaces (BCIs) is promising in interacting with machines through electroencephalogram (EEG) signal. The compact end-to-end neural network model for generalized BCIs, EEGNet, has been implemented in hardware to get near sensor intelligence, but without enough efficiency. To utilize EEGNet in low-power wearable device for long-term use, this paper proposes an efficient EEGNet inference accelerator. Firstly, the EEGNet model is compressed by embedded channel selection, normalization merging, and product quantization. The customized accelerator based on the compressed model is then designed. The multilayer convolutions are achieved by reusing multiplying-accumulators and processing elements (PEs) to minimize area of logic circuits, and the weights and intermediate results are quantized to minimize memory sizes. The PEs are clock-gated to save power. Experimental results in FPGA on three datasets show the good generalizing ability of the proposed design across three BCI diagrams, which only consumes 3.31% area and 1.35% power compared to the one-to-one parallel design. The speedup factors of 1.4, 3.5, and 3.7 are achieved by embedded channel selection with negligible loss of accuracy (-0.80%). The presented accelerator is also synthesized in 65 nm CMOS low power (LP) process and consumes 0.23M gates, 24.4 ms/inference, 0.267 mJ/inference, which is 87.22% more efficient than the implementation of EEGNet in a RISC-V MCU realized in 40 nm CMOS LP process in terms of area, and 20.77% more efficient in terms of energy efficiency on BCIC-IV-2a dataset.

The ECG classification processor is a key component in wearable intelligent ECG monitoring devices which monitor the ECG signals in real time and detect the abnormality automatically. The state-of-the-art ECG classification processors for wearable intelligent ECG monitoring devices are faced with two challenges, including ultra-low energy consumption demand and high classification accuracy demand against patient-to-patient variability. To address the above two challenges, in this work, an ultra-energy-efficient ECG classification processor with high classification accuracy is proposed. Several design techniques have been proposed, including a reconfigurable SNN/ANN inference architecture for reducing energy consumption while maintaining classification accuracy, a reconfigurable on-chip learning architecture for improving the classification accuracy against patent-to-patient variability, and a dual-purpose binary encoding scheme of ECG heartbeats for further reducing the energy consumption. Fabricated with a 28nm CMOS technology, the proposed design consumes extremely low classification energy (0.3μJ) while achieving high classification accuracy (97.36%) against patient-to-patient variability, outperforming several state-of-the-art designs.

This paper presents a 14-b 20-MS/s energy-efficient SAR ADC in 65-nm CMOS technology for portable medical ultrasound systems. To break the limitation of the ADC linearity on the DAC size in a SAR ADC, a background mismatch calibration technique is employed. As a result, the thermal noise will be the major constraint for the DAC size. In addition, a compact noise-reduction technique is proposed to alleviate the adverse impact of the input-referred comparator noise on the effective resolution. Moreover, a 2.5-V on-chip LDO, which serves as the reference generator for the ADC core, is also integrated to guarantee the reference accuracy and to suppress the supply noise. To reduce the capacitive load of the comparator and boost the comparison speed, a low fan-in SAR logic is also designed. With the proposed mismatch calibration technique and the noise-reduction technique activated, measured results indicate that the peak signal-to-noise-and-distortion ratio (SNDR) and the spurious-free dynamic range (SFDR) achieve 78.8 dB and 95.4 dB, respectively. At 20 MS/s, the ADC consumes 6.8mW from its 1.2 V/3.3V supplies in total, leading to an SNDR-based Schreier FOM of 170.5 dB at Nyquist. The active area of the ADC is 450 × 540μm 2 .

Recent research has advanced two degree-of-freedom (DoF), simultaneous, independent and proportional control of hand-wrist prostheses using surface electromyogram signals from remnant muscles as the control input. We evaluated two such regression-based controllers, along with conventional, sequential two-site control with co-contraction mode switching (SeqCon), in box-block, refined-clothespin and door-knob tasks, on 10 able-bodied and 4 limb-absent subjects. Subjects operated a commercial hand and wrist using a socket bypass harness. One 2-DoF controller (DirCon) related the intuitive hand actions of open-close and pronation-supination to the associated prosthesis hand-wrist actions, respectively. The other (MapCon) mapped myoelectrically more distinct, but less intuitive, actions of wrist flexion-extension and ulnar-radial deviation. Each 2-DoF controller was calibrated from separate 90 s calibration contractions. SeqCon performed better statistically than MapCon in the predominantly 1-DoF box-block task (>20 blocks/minute vs. 8-18 blocks/minute, on average). In this task, SeqCon likely benefited from an ability to easily focus on 1-DoF and not inadvertently trigger co-contraction for mode switching. The remaining two tasks require 2-DoFs, and both 2-DoF controllers each performed better (factor of 2-4) than SeqCon. We also compared the use of 12 vs. 6 optimally-selected EMG electrodes as inputs, finding no statistical difference. Overall, we provide further evidence of the benefits of regression-based EMG prosthesis control of 2-DoFs in the hand-wrist.

System identification models relating forearm electromyogram (EMG) signals to phantom wrist radial-ulnar deviation force, pronation-supination moment and/or hand open-close force (EMG-force) are hampered by lack of supervised force/moment output signals in limb-absent subjects. In 12 able-bodied and 7 unilateral transradial limb-absent subjects, we studied three alternative supervised output sources in one degree of freedom (DoF) and 2-DoF target tracking tasks: (1) bilateral tracking with force feedback from the contralateral side (non-dominant for able-bodied/ sound for limb-absent subjects) with the contralateral force as the output, (2) bilateral tracking with force feedback from the contralateral side with the target as the output, and (3) dominant/limb-absent side unilateral target tracking without feedback and the target used as the output. "Best-case" EMG-force errors averaged ~ 10% of maximum voluntary contraction (MVC) when able-bodied subjects' dominant limb produced unilateral force/moment with feedback. When either bilateral tracking source was used as the model output, statistically larger errors of 12-16 %MVC resulted. The no-feedback alternative produced errors of 25-30 %MVC, which was nearly half the tested force range of ± 30 %MVC. Therefore, the no-feedback model output was not acceptable. We found little performance variation between DoFs. Many subjects struggled to perform 2-DoF target tracking.

Typical electromyogram (EMG) processors estimate EMG signal standard deviation (EMG σ ) via moving average root mean square (RMS) or mean absolute value (MAV) filters, whose outputs are used in force estimation, prosthesis/orthosis control, etc. In the inevitable presence of additive measurement noise, some processors subtract the noise standard deviation from EMG RMS (or MAV). Others compute a root difference of squares (RDS)-subtract the noise variance from the square of EMG RMS (or MAV), all followed by taking the square root. Herein, we model EMG as an amplitude-modulated random process in additive measurement noise. Assuming a Gaussian (or, separately, Laplacian) distribution, we derive analytically that the maximum likelihood estimate of EMG σ requires RDS processing. Whenever that subtraction would provide a negative-valued result, we show that EMG σ should be set to zero. Our theoretical models further show that during rest, approximately 50% of EMG σ estimates are non-zero. This result is problematic when EMG σ is used for real-time control, explaining the common use of additional thresholding. We tested our model results experimentally using biceps and triceps EMG from 64 subjects. Experimental results closely followed the Gaussian model. We conclude that EMG processors should use RDS processing and not noise standard deviation subtraction.

Endoscopic integrated photoacoustic and ultrasound imaging has the potential for early detection of the cancer in the gastrointestinal tract. Currently, slow imaging speed is one of the limitations for clinical translation. Here, we developed a high speed integrated endoscopic PA and US imaging system, which is able to perform PA and US imaging simultaneously up to 50 frames per second. Using this system, the architectural morphology and vasculature of the rectum wall were visualized from a Sprague Dawley rat in-vivo .

To facilitate stretch reflex onset (SRO) detection and improve accuracy and reliability of spasticity assessment in clinical settings, a new method to measure dynamic stretch reflex threshold (DSRT) based on Hilbert-Huang transform marginal spectrum entropy (HMSEN) of surface electromyography (sEMG) signals and a portable system to quantify modified Ashworth scale (MAS) for spasticity assessment were developed. The sEMG signals were divided into frames using a fixed-length sliding window, and the HMSEN of each frame was calculated. An adaptive threshold was set to measure the DSRT. The HMSEN based method can quantify muscle activity through time-frequency and nonlinear dynamics analysis, therefore providing deeper insight about the spastic muscle mechanisms during stretching and a reliable SRO detection method. Experimental results revealed that the HMSEN based method could reliably detect the SRO and measure the DSRT (recognition rate: 95.45%), and could achieve improved performance over the time-domain based method. There was a strong correlation ( to -0.900) between the MAS scores and the DSRT index, and the test-retest reliability was high. Additionally, limitations of the MAS were analyzed. This paper indicates that the presented framework can provide a promising tool to measure DSRT and a clinical quantitative approach for spasticity assessment.

The extremely low tissue contrast in white matter during an infant's isointense stage (6-8 months) of brain development presents major difficulty when segmenting brain image regions for analysis. We sought to develop a label-fusion-aided deep-learning approach for automatically segmenting isointense infant brain images into white matter, gray matter and cerebrospinal fluid using T1- and T2-weighted magnetic resonance images. A key idea of our approach is to apply the fully convolutional neural network (FCNN) to individual brain regions determined by a traditional registration-based segmentation method instead of training a single model for the whole brain. This provides more refined segmentation results by capturing more region-specific features. We show that this method outperforms traditional joint label fusion and FCNN-only methods in terms of Dice coefficients using the dataset from iSEG MICCAI Grand Challenge 2017.