Epilepsy is a neurological disorder characterized by different symptoms, including seizures, stiffness of muscles, uncontrollable movement of the body, and multiple psychological abnormalities. Electroencephalogram (EEG) is extensively used in the diagnosis of epilepsy. The presence of abnormal wave complexes observed in the EEG signal helps in the diagnosis of epilepsy. There are multiple databases containing EEG signals from epileptic patients, which are used in developing support systems for neurologists to augment the diagnosis or prediction of epilepsy. However, few epilepsy databases are specific to the South-Asian region. Of these databases, no continuous recording of EEG signals with multiple wave complexes in unipolar and bipolar EEG signals is provided. This study aims to publish the first South-Asian database containing continuous EEG signals from chronic epileptic patients with multiple wave complexes in unipolar and bipolar formats. Moreover, apart from the wave complexes, the EEG recordings have also been marked for artifacts which may be used to train the system for automatic rejection of the artifact noise from the EEG epileptic signals.

Current electroencephalogram (EEG) measuring systems are bulky, impose constraints on patients, and require pre and post-measuring procedures. Usually, the EEG systems use either wet or dry EEG sensors, with the former suffers from skin preparation, the issues of adhesive conductive gels, and one-time usability whereas the latter causes skin irritation, abrasion, and pain upon pressure. Hence, these sensors are not suitable for long-term measurements. This paper presents a novel, wireless, behind-the-ear wearable EEG acquisition device that incorporates flexible dry EEG sensors. Silver ink-printed flexible sensors are fabricated using screen printing to overcome the above-mentioned drawbacks and limitations of conventional EEG sensors. The flexible sensors form a capacitive link with the skin via an adhesive layer between the sensor and the person's skin and are capable of acquiring the EEG without any skin preparation or gel. The performance of the printed flexible EEG sensors is tested by comparing them with the standard Ag/AgCl pre-gelled sensors. The alpha wave test and evoked potential EEG test are also performed for verification. The proposed device has a small form factor similar to a hearing aid and an in-house configurable Analog Front End (AFE) and Digital Back End (DBE) Processor and is capable of acquiring continuous EEG for a longer duration in a user-friendly and socially discrete manner.

Epileptic patients ’ quality of life can be significantly improved by epileptic seizure prediction based on scalp electroencephalogram (EEG). With the advancement of brain e-health technologies, there is an essential need for a method that accurately predicts seizures while running on computing platforms with very low computing resources. Moreover, existing methods do not provide EEG analysis on an individual channel basis to identify the abnormalities in the data. In order to address this issue, we propose an efficient framework for patient-specific seizure prediction. A hybrid model comprising of a shallow autoencoder (AE) with only one hidden layer and a support vector machine (SVM) classifier has been developed. Both multi-channel and single channel EEG signal processing schemes have been developed. Generating a lower dimensional sparse signal with AE in the first stage and classifying the signal using SVM in the second stage are the two stages that the model separates into when processing EEG data. We initially train the AE to provide an optimum sparse signal and then use this sparse signal as input for an SVM classifier to categorize the EEG data. Using the 10-fold cross validation strategy, the proposed model tests 13 patients from the CHB-MIT dataset and achieves an average sensitivity of 98% and an average area under the curve (AUC) of 99%. We have compared our hybrid approach ’s performance with both deep learning models and traditional techniques. The proposed methodology outperforms state of the art seizure prediction methods, demonstrating its effectiveness.

Blood pressure (BP) is considered one of the key vital signs that provide valuable medical information about cardiovascular activity. Conventionally, cuff-based devices are used to measure BP which limits their usage for continuous monitoring. This paper presents a cuff-less BP estimation processor using photoplethysmography (PPG) signals with a Deep Neural Network (DNN). Spectral and temporal features are extracted from the PPG signals and then used to train and evaluate the machine learning (ML) algorithms. The proposed algorithm is evaluated using the MIMIC II database for systolic blood pressure (SBP) and diastolic blood pressure (SBP) estimation. The proposed BP estimation processor is implemented using a 180nm CMOS process with an area of 3.45mm2and consumes $73 \mu \mathrm{W}$. It achieves a mean absolute error in systolic BP of $0.0657 \pm 4.7$ mmHg and diastolic BP of $0.792 \pm 4.61$ mmHg which outperforms the state-of-the-art BP estimation algorithms.

An electroencephalogram (EEG) based non-invasive 2-channel System on Chip (SoC) is presented to detect and report the seizure event of the epileptic patient. The SoC incorporates an area and power-efficient dual-channel analog front-end (AFE) and machine learning-based differential difference approximate computing seizure detection ($\text{D}^{2}$ACSD) processor. The $\text{D}^{2}$ACSD processor integrates approximate computing feature extraction and fixed-point linear support vector machine (LSVM) classifier to minimize the area-and-power utilization. The AFE comprises of two duty-cycled resistive MOSFET (DCRM) capacitively coupled instrumentation amplifier ($\text{C}^{2}$IA), a programmable gain amplifier, and multiplexed SAR-ADC. The DCRM-C2IA utilizes proposed DCRM technique to boost the equivalent resistance of the integrator of the DC servo loop. The 5m$\text{m}^{2}$SoC is implemented in 0.18$\mu$m, CMOS process while achieving an average accuracy of 89.19%, sensitivity 92.18% and specificity 89.13% for the random and block-wise splitting of data in train/test sets. The implemented DCRM-C2IA achieves an integrated noise of 0.80$\mu$Vrms over 0.5-100Hz frequency band. The realized system consumes $2.7\mu \text{J}/$classification to continuously detect seizure onset for timely suppression.

Health problems related to stress are increasing globally and significantly affect the mental health and quality of life of human beings. Continuous suffering from stress may lead to serious psychological and physical health problems. But still, no effective and reliable stress detection methods are available. In this paper, a novel wearable device is presented to measure electroencephalogram (EEG) and electrocardiogram (ECG) simultaneously in a non-invasive approach. This system includes an analog front end (AFE) integrated with a machine learning-based digital backend (DBE) processor for mental stress prediction using only 3 electrodes. A PCB prototype is developed using the commercial off-the-shelf components. The developed prototype shows excellent noise performance of $0.1\mu V_{rms}$ and predicts the mental stress with a classification accuracy of 92.7%. The proposed system is lightweight and easily wearable (behind the ear). The data is acquired from 25 participants for different stress scenarios including the Arithmetic Test and Stroop Color Word Test. Different EEG and ECG based features combinations are used for the classification of stress conditions using a shallow neural network (SNN) classifier.

Tone mapping is necessary for Low Dynamic Range (LDR) devices to display High Dynamic Range (HDR) images and videos. Multiple video Tone Mapping Operators (vTMOs) have been devised for HDR videos. The majority of vTMOs apply an image TMO to each video frame. This is followed by pre/post filtering to ensure temporal coherence. However, this destroys the natural temporal variation in intensity that is inherent in a changing scene. Furthermore, in these methods computational complexity of an image TMO is scaled up in proportion to the number of frames. We propose an efficient method to extend an image TMO to video TMO. The proposed method is general and takes care of temporally coherent intensity variation between frames while addressing the well-known issue of flickering in tone mapped video. Additionally, it lowers computational complexity as a new tone mapping curve (TMC) is not generated on per frame basis. The proposed vTMO can be used to extend any state-of-the-art global TMO that is deemed to generate a TMC. Fresh TMC is generated only when a hard cut in video is detected or the global change in illumination in HDR video becomes large enough. Visual comparisons and objective evaluations with three well known image TMOs demonstrate that the suggested extension method generates high quality LDR video generally at low computational cost when compared to existing vTMOs. Further work on efficient implementation of embedding an image TMO in our vTMO algorithm is expected to yield even better computational efficiency.

A charge-recycled amplifier-less current-to-digital converter (CDC) is presented in this paper. The highly digital CDC is based on ultra-low-power subthreshold operation, amplifier-free design, scalable current range, and high-resolution digital circuits. It is realized using a mixed-signal approach by utilizing an analog comparator, 6-bits current steering DAC, an implicit passive integrator, and digital signal conditioning circuits. The proposed converter can be operated in tri-modes to support multiple current sensing applications, i.e. Mode1: current range (CR) from 10nA to 400nA, Mode 2: CR from 0.5μA to 5μA, and Mode3: CR from 2.5μA to 42.5μA. The proposed CDC is implemented in 180nm CMOS process with an active area of 0.54 mm2while consuming 11.2μW. It achieves a 77.8dB dynamic range at a bandwidth of 300Hz from 1V supply with an FoM of 114.2fJ/step.

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.

Accurate monitoring of the depth of anesthesia (DoA) is essential for intraoperative and postoperative patient's health. Commercially available electroencephalograph (EEG)-based DoA monitors are recommended only for certain anesthetic drugs and specific age-group patients. This paper presents a machine learning classification processor for accurate DoA estimation irrespective of the patient's age and anesthetic drug. The classification is solely based on six features extracted from EEG signal, i.e., spectral edge frequency (SEF), beta ratio, and four bands of spectral energy (FBSE). A machine learning fine decision tree classifier is adopted to achieve a four-class DoA classification (deep, moderate, and light DoA versus awake state). The feature selection and the classification processor are optimized to achieve the highest classification accuracy for the state of moderate anesthesia required for the surgical operations. The proposed 256-point fast Fourier transform accelerator is implemented to realize SEF, beta ratio, and FBSE that enables minimal latency and high accuracy feature extraction. The proposed DoA processor is implemented using a 65 nm CMOS technology and experimentally verified using field programming gate array (FPGA) based on the EEG recordings of 75 patients undergoing elective surgery with different types of anesthetic agents. The processor achieves an average accuracy of 92.2% for all DoA states, with a latency of 1s The 0.09 mm2DoA processor consumes 140nJ/classification.

Autism Spectrum Disorder (ASD) is a neurodevelopment disorder that affects children's development and can lead to handicap life if remain untreated. Scalp Electroencephalography (EEG) data can be used as a biomarker to characterize the human emotions on the valence-arousal scale. This work presents a machine learning patient-specific emotion detection (PSED) classification processor based on an eight-channel EEG signal. The proposed PSED classification processor integrates a hardware-efficient feature extraction engine and patient-specific support vector machine (SVM) classifier to discriminate the emotions in real-time. To utilize minimal hardware resources a hardware realizable feature set comprising of power spectral density (PSD), an absolute difference of inter-hemispheric power asymmetry (IHPD), and the scaled inter-hemispheric power asymmetry ratio (SIHPR) of eight electrode pairs are evaluated. To avoid high overhead of area and power consumption for an integer divider for SIHPR; simple LUT based divider is proposed that calculates the approximated value of SIHPR with a minimal overhead of 64 Bytes. The classification is performed using a Linear SVM and resulted in an accuracy of 63% and 60% for valence and arousal, respectively, based on the database for emotion analysis using physiological signals (DEAP). The PSED processor is synthesized using a 65nm CMOS technology with an overall energy efficiency of 10uJ/classification.

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.

We consider a two-user non-orthogonal-multiple-access (NOMA) communication channel with an iterative multi-user receiver. It is known that NOMA provides performance gains over conventional orthogonal-multiple-access, but it comes at the cost of increased decoder complexity. Moreover, the decoder complexity varies with fraction of duration in which the users' transmissions overlap using NOMA. For this scheme, we present a hardware-friendly implementation of an LDPC-code-based multi-user-detector (MUD). We also present a MATLAB-based high-level design-automation tool that generates hardware descriptions of different NOMA-receivers. Results indicate that the simplified MUD design results in significant area and power savings with negligible impact on performance.

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.

Classification of EEG under wearable environment faces many challenges including motion artifact, electrode DC offset, noise and limited available energy source. This paper describes the design consideration of a multi-channel machine-learning based EEG classification and recording processors for wearable form-factor sensors. The goal is to optimize the detection performance while balancing the analog and digital signal processing to optimize its energy consumption. On-chip classification significantly helps achieving energy-efficiency by reducing the communication overhead of the data. With epileptic seizure detection and recording system examples, we start from choosing number of channels, the sampling rate, and how to effectively extract features out of the down-sampled data. After that, classification algorithms are also discussed in detail. When verified with the Children's Hospital Boston-Massachusetts Institute of Technology (CHB-MIT) EEG database, based on Repeated Random Sub-Sampling validation, the seizure detection sensitivity and specificity of the Non-Linear SVM are improved by 12.4%P and 3.56%P, respectively, compared to the Linear-SVM. The LSVM and NLSVM processors are fabricated in 0.18μm 1P6M CMOS and consume 1.52μJ/classification and 1.34μJ/classification, respectively. Finally, the on-chip memory requirements for storing the raw seizure data is discussed.

This paper presents the design of an area- and energy-efficient closed-loop machine learning-based patient-specific seizure onset and termination detection algorithm, and its on-chip hardware implementation. Application- and scenario-based tradeoffs are compared and reviewed for seizure detection and suppression algorithm and system which comprises electroencephalography (EEG) data acquisition, feature extraction, classification, and stimulation. Support vector machine achieves a good tradeoff among power, area, patient specificity, latency, and classification accuracy for long-term monitoring of patients with limited training seizure patterns. Design challenges of EEG data acquisition on a multichannel wearable environment for a patch-type sensor are also discussed in detail. Dual-detector architecture incorporates two area-efficient linear support vector machine classifiers along with a weight-and-average algorithm to target high sensitivity and good specificity at once. On-chip implementation issues for a patient-specific transcranial electrical stimulation are also discussed. The system design is verified using CHB-MIT EEG database [1] with a comprehensive measurement criteria which achieves high sensitivity and specificity of 95.1% and 96.2%, respectively, with a small latency of 1 s. It also achieves seizure onset and termination detection delay of 2.98 and 3.82 s, respectively, with seizure length estimation error of 4.07 s.

This paper presents an 8-channel electroencephalograph (EEG) classification processor for seizure detection and recording. To integrate 8 channels, an area- and energy-efficient filter architecture using Distributed Quad-LUT (DQ-LUT) is proposed, which reduces area by 64.2% with minimal overhead in power-delay product. The on-chip patient specific classification with a Linear Support-Vector Machine (SVM) results in 82.7% seizure detection accuracy with a 2 second latency using the CHB-MIT EEG database [1]. The overall energy efficiency is measured as 1.52μJ/classification while operating at 8 channel mode.