Search

Long-term, continuous monitoring of human brain activity with closed-loop precise neurostimulation can potentially help to treat conditions such as epilepsy and Parkinson. These implantable devices are used to sense the brain signal, detect an abnormality, and stimulate once the abnormal activity is detected to mitigate the adverse effect. The target is to achieve the maximum efficacy while balancing the signal acquisition and intelligent processing to ensure minimize battery replacement frequency. This paper presents the design of the implantable Electrocorticography (ECoG) based system for intractable epileptic seizure treatment. The challenges, design choices, and trade-offs, related to the neurological disorder in the implantable environment are discussed. A multi-channel ultra-low-power instrumentation amplifier (IA) with digital electrode offset rejection loop (EORL) with a cutoff frequency of < 0.5Hz to mitigate the electrode offset effect with a fast settling of < 0.1 Sec to ensure real-time recording. The implantable system is realized in 180nm CMOS process to ensure area-and-power efficient design.

An EEG-based noninvasive neuro-feedback SoC for emotion classification of Autistic children is presented. The AFE comprises two entirely shared EEG-channels using sampling capacitors to reduce the area by 30% and achieve an overall integrated input-referred noise of 0.55µ VRMS with cross-talk of - 79dB. The 4-layers Deep Neural Network (DNN) classifier is integrated on-sensor to classify (4 emotions) with >85% accuracy. The 16mm2 SoC in 0.18um CMOS consumes 10.13µJ/classification for 2 channels.

Sleep stages classification is an effective tool for the diagnosis and treatment of neurodegenerative disorders. This paper presents the first non-invasive electroencephalograph (EEG)-based processor for classifying all the sleep stages implemented on hardware. It utilizes a single EEG channel and multi-machine-learning classifiers to form a home-based polysomnography. These multiple one-vs-one binary Linear Support Vector Machine (LSVM) classifiers are combined to classify all the sleep stages using two features only. To implement the desired features efficiently on hardware, an exponent-eliminate (EE) Split-Radix 256-point FFT is proposed that decreases the area by 60% compared to the conventional design by avoiding the majority of complex floating-point multiplications and divisions. The proposed all sleep stages classification system is implemented using 180nm CMOS process and experimentally verified using FPGA based on the EEG recordings of 197 patients from Physionet Sleep database. It utilizes a miniaturized active area of 0.32mm2 and achieves a Cohen Kappa score of 0.847 while consuming 0.7µJ/classification.

An Electrocardiography (ECG) based processor for eight Cardiac arrhythmias (CA) detection with smart priority logic is presented to minimize the false alarms. The processor utilizes a Multi-Level Linear Support Vector Machine (ML-LSVM) classifiers with one-vs-all approach to distinguish the different CAs. The classification is solely based on 5 features including R-wave, S-wave, T-wave, R-R interval and Q-S interval. The processor employs a priority logic to prioritize the detected conditions if more than one condition are detected. The system is implemented using CMOS 180nm with an area of 0.18mm2 and validated using 83 patient’s recordings from Physionet Arrhythmia Database and Creighton University Database. The proposed processor consumes 0.91uW with an average classification accuracy of 98.5% while reducing the false alarms by 99%, which is 30% superior performance compared to conventional systems.

Peripheral neuropathy is a condition that affects the normal activity of the nerves of the peripheral nervous system. Patients with peripheral neuropathy may have tingling, numbness, unusual sensations, weakness, or burning pain in the affected area. The prevalence of this ailment ranges from 50% in Turkey, 22-600% in the Middle East, and 69% in Pakistan. Our aim is to devise a simple and reliable method to detect neuropathy. The Electromyography (EMG) signal can be used to analyze medical abnormalities and to analyze the bio medics of the human body. The signal carries valuable information regarding the nerve system. The electromyogram will be used to record the electrical activity through the surface electrodes applied to the skin. This work aims to extract the EMG signal through various modules: pre-amplification, frequency response analysis and processing of the signal extracted from both healthy and the ones with the tremor. At the end, we will be able to differentiate a healthy person from the one suffering from the ailment. It is supposed that the analysis of EMG signal will have wider application and research of tremor detection.

Continuous glucose monitoring is essential for patients to avoid complications of both hypoglycemia and hyperglycemia. This paper presents a novel non-invasive continuous blood glucose monitoring system based on a single wavelength near-infrared (NIR) spectroscopy. The analog frontend of the system is designed with a single NIR LED to record the Photoplethysmographic (PPG) signal from the fingertip with motion artifacts removal and a bias current rejection up to 20uA. The proposed digital backend extracts 10 discriminating features from the PPG signal to predict the blood glucose level using (Exponential Gaussian Process) machine learning regression. To realize the feature extraction on FPGA, a novel two-dimensional structure of 256-point Fast Fourier Transform (FFT) is implemented which achieves a 47% reduction in complex multiplications compared to the conventional Radix-2 algorithm. The performance of the proposed system is validated using 200 patients PPG recordings and glucose levels measured via a commercial glucometer. It successfully predicts the glucose level with a mean absolute relative difference (mARD) of 8.97%.

The increasing prevalence of neurodegenerative diseases (NDDs) impose substantial medical and public health burdens on populations throughout the world. NDDs are chronic diseases that affect the human central nervous system causing loss of neurons within the brain and/or spinal cord. This causes deterioration in movement and mental functioning of the patients. The current medications for this group of disorders are limited and aim to treat the symptoms only. A better understanding of the mechanisms underlying neurodegeneration should lead to more effective, disease-modifying treatments in the future. Continuous assessment of NDD patients is a key element of future care and treatment. This contribution proposes a wearable NDD detection system based on patient’s gait dynamics using an unobtrusive force resistive sensor embedded in patient’s shoe. The NDD classification is based on 3 fundamental gait features: stride time, stride time’s fluctuation and the autocorrelation decay factor. It is designed to discriminate between healthy subjects and NDD patients and moreover identify the NDD type: (Huntington’s disease (HD), Parkinson Disease (PD), and Amyotrophic Lateral Sclerosis (ALS)). The proposed NDD classification algorithm is implemented on FPGA and verified experimentally using Gait Dynamics dataset from Physionet. It offers a classification accuracy of 93.8%, 89.1%, 94% and 93.3%, for ALS, HD, PD, and healthy person, respectively, from a total set of 64 subjects.

A non-machine learning patient-specific shockable cardiac arrhythmia (SCA) classification processor based on single lead electrocardiogram (ECG) is presented. The proposed SCA detection processor integrates a hardware-efficient reduced-set-of-five (RSF5) feature extraction engine to extract SCA and non-SCA, self-adaptive patient-specific threshold engine for the peak and interval detection from the ECG, and simplified decision logic to discriminate the arrhythmia in real-time. The SCAD processor consumes 0.89pJ/classification while classifying with an average sensitivity, and specificity of 98.66%, and 99.75%, respectively.

An electroencephalograph (EEG) based classification processor for the depth of Anesthesia (DoA) during the intraoperative procedure is presented. To enable a DoA to monitor the correct estimation across a range of patients, a novel feature extraction along with machine learning processor is utilized. The decisions are solely based on seven features extracted from EEG along with the EMG signal for motion artifacts rejection. To extract the features efficiently on hardware, a 128-point FFT is proposed that achieves an area reduction and energy/FFT-operation by 39% and 58%, respectively, compared to the conventional. A simple decision tree is used to perform a multiclass DoA classification. The system is synthesized using a 65nm process and experimental verification is done using FPGA based on the subset of patients from the University of Queensland Vital Signs. The proposed patient-specific DoA classification processor achieves a classification accuracy of 79%.

Cardiac arrhythmia (CA) is one of the most serious heart diseases that lead to a very large number of annual casualties around the world. The traditional electrocardiography (ECG) devices usually fail to capture arrhythmia symptoms during patients' hospital visits due to their recurrent nature. This paper presents a wearable long-term single-lead ECG processor for the CA detection at an early stage. To achieve on-sensor integration and long-term continuous monitoring, an ultra-low complexity feature extraction engine using reduced feature set of four (RFS4) is proposed. It reduces the area by >25% compared to the conventional QRS complex detection algorithms without compromising the accuracy. Moreover, RFS4 eliminates the need for complex machine learning decision logic for the detection of premature ventricular contraction (PVC) and nonsustained ventricular tachycardia (NVT). To ensure correct functional verification, the proposed system is implemented on FPGA and tested using the MIT-BIH ECG arrhythmia database. It achieves a sensitivity and specificity of 94.64% and 99.41%, respectively. The proposed processor is also synthesized using 0.18um CMOS technology with an overall energy efficiency of 139 nJ/detection.

An electroencephalograph (EEG) based classification processor for Random Eye-movement (REM) and Non-REM (NREM) sleep detection for Alzheimer patients is presented. To enable a home-based wearable polysomnography, a single-channel-EEG, with the novel feature extraction (FE) approach and the machine-learning classifier is utilized. The 4-dimenional Feature Vector (FV) is based on the spectral edge frequency (SEF), low-to-high spectral content, and rapid-eye-movement detection. To utilize SEF efficiently on hardware, a 16-point FFT is proposed that achieves an area reduction of 43% compared to the conventional with minimal overhead in the precision loss. A Linear Support Vector Machine (LSVM) is utilized to decide between REM and NREM based on the FV. The system is synthesized using 0.18μm process and experimental verification is done using FPGA based on the subset of patients from the PhysioNet Sleep database. The proposed classification processor achieves energy-efficiency of 0.21J/classification with a sensitivity and specificity of 89.8% and 93.6% respectively.

Neurodegenerative disorders (NDDs) are chronic diseases of the human central nervous system that cause degradation in mobility and cognitive functioning. Continuous assessment of gait for patients with NDDs is a crucial element of future care and treatment. This paper presents a wearable NDD detection system that monitors the person's gait and infers 3 key gait features: stride time, its fluctuation and autocorrelation decay factor based on data extracted from an unobtrusive force resistive sensor embedded in patient's shoe. It is designed to distinguish between different NDDs: (Huntington's disease (HD), Parkinson Disease (PD), and Amyotrophic Lateral Sclerosis (ALS)) and healthy individuals using only 3 features. The proposed NDD classification algorithm is verified experimentally using a full FPGA implementation with patients' recordings from Physionet Gait Dynamics data set. It achieves a classification accuracy of 93.8%, 89.1%, 94% and 93.3%, for ALS, HD, PD, and healthy person, respectively, from a total set of 64 subjects.

This paper presents a 2-channel electroencephalograph (EEG) based seizure detection accelerator suitable for long-term continuous monitoring of patients suffering from the Grand-mal seizure disorder. The implementation is based on the novel slope based detection (SBD) algorithm to achieve start and end of seizure detection. The proposed SBD algorithm is verified experimentally using a full FPGA implementation with patients' recordings from Physionet Children Hospital Boston (CHB)-MIT EEG database with real-time seizure, information display on the Android phone through a low-power Bluetooth link. The patient-specific detection with specific threshold results in sensitivity, specificity, system latency, and detection latency of 91.2%, 93.6%, 0.5s, and 29.25 s, respectively, using the CHB-MIT EEG database.