SPIRIT: Low Power Seizure Prediction using Unsupervised Online-Learning and Zoom Analog Frontends

Early prediction of seizures and timely interventions are vital for improving patients' quality of life. While seizure prediction has been shown in software-based implementations, to enable timely warnings of upcoming seizures, prediction must be done on an edge device to reduce latency. Ideally, such devices must also be low-power and track long-term drifts to minimize maintenance from the user. This work presents SPIRIT: Stochastic-gradient-descent-based Predictor with Integrated Retraining and In situ accuracy Tuning. SPIRIT is a complete system-on-a-chip (SoC) integrating an unsupervised online-learning seizure prediction classifier with eight 14.4 uW, 0.057 mm2, 90.5 dB dynamic range, Zoom Analog Frontends. SPIRIT achieves, on average, 97.5%/96.2% sensitivity/specificity respectively, predicting seizures an average of 8.4 minutes before they occur. Through its online learning algorithm, prediction accuracy improves by up to 15%, and prediction times extend by up to 7x, without any external intervention. Its classifier consumes 17.2 uW and occupies 0.14 mm2, the lowest reported for a prediction classifier by >134x in power and >5x in area. SPIRIT is also at least 5.6x more energy efficient than the state-of-the-art.

Towards EMG-to-Speech with a Necklace Form Factor

Electrodes for decoding speech from electromyography (EMG) are typically placed on the face, requiring adhesives that are inconvenient and skin-irritating if used regularly. We explore a different device form factor, where dry electrodes are placed around the neck instead. 11-word, multi-speaker voiced EMG classifiers trained on data recorded with this device achieve 92.7% accuracy. Ablation studies reveal the importance of having more than two electrodes on the neck, and phonological analyses reveal similar classification confusions between neck-only and neck-and-face form factors. Finally, speech-EMG correlation experiments demonstrate a linear relationship between many EMG spectrogram frequency bins and self-supervised speech representation dimensions.

A Wireless Ear EEG Drowsiness Monitor

Wireless, neural wearables can enable life-saving drowsiness, cognitive, and health monitoring for heavy machinery operators, pilots, and drivers. While existing systems use in-cabin sensors to alert operators before accidents, wearables may enable monitoring across many user environments. Current neural wearables are promising but limited by consumable electrodes and bulky, wired electronics. To improve neural wearable usability, scalability, and enable discreet use in daily and itinerant environments, this work showcases the end-to-end design of the first wireless, in-ear, dry-electrode drowsiness monitoring platform. The proposed platform integrates additive manufacturing processes for gold-plated dry electrodes, user-generic earpiece designs, wireless electronics, and low-complexity machine learning algorithms. To evaluate the platform, thirty-five hours of ExG data were recorded across nine subjects performing repetitive drowsiness-inducing tasks. The data was used to train three, offline classifier models (logistic regression, support vector machine, and random forest) and evaluated with three training regimes (user-specific, leave-one-trial-out, and leave-one-user-out). The support vector machine classifier achieved an average accuracy of 93.2% while evaluating users it has seen before and 93.3% when evaluating a never-before-seen user. These results demonstrate for the first time that dry, 3D printed, user-generic electrodes can be used with wireless electronics to rapidly prototype wearable systems and achieve comparable average accuracy (>90%) to existing state-of-the-art in-ear and scalp ExG systems that utilize wet electrodes and wired, benchtop electronics. Further, this work demonstrates the feasibility of using population-trained machine learning models in future, wearable ear ExG applications focused on cognitive health and wellness tracking.