arXiv subjects

Maria Nyamukuru

Publications and source records attributed to Maria Nyamukuru.

3 recordsLinked to original sources

arXiv2022

Analog Gated Recurrent Neural Network for Detecting Chewing Events

We present a novel gated recurrent neural network to detect when a person is chewing on food. We implemented the neural network as a custom analog integrated circuit in a 0.18 um CMOS technology. The neural network was trained on 6.4 hours of data collected from a contact microphone that was mounted on volunteers' mastoid bones. When tested on 1.6 hours of previously-unseen data, the neural network identified chewing events at a 24-second time resolution. It achieved a recall of 91% and an F1-score of 94% while consuming 1.1 uW of power. A system for detecting whole eating episodes -- like meals and snacks -- that is based on the novel analog neural network consumes an estimated 18.8uW of power.

cs.LG↗

arXiv2021

Extracting Fractional Inspiratory Time from Electrocardiograms

Non-invasive at-home monitoring of lung and lung airways health enables the early detection and tracking of respiratory diseases like asthma and chronic obstructive pulmonary disease (COPD). Various proposed approaches estimate the respiratory rate and extract the respiratory waveform from an electrocardiogram (ECG) signal as a way to discreetly monitor lung health. Unfortunately, these approaches fail to accurately capture the respiratory cycle phase features, resulting in a non-specific, incomplete picture of lung health. This paper introduces an algorithm to extract more respiratory information from the ECG signal by framing the problem as a binary segmentation task. In addition to respiratory rate (RR), the algorithm derives the fractional inspiratory time (FIT), a direct measure of airway obstruction derived from respiratory phase information. The algorithm is based on a gated recurrent neural network that infers vital respiratory information from a single-lead ECG signal. We measure our algorithm's performance on 5 subjects from the MIMIC dataset and 5 subjects from the CEBS database. Our algorithm maintains exceptional performance in estimating the respiratory rate and outperforms current algorithms that extract the respiratory cycle phases and FIT/ inspiratory:expiratory ratio (IER). Our algorithm reports a root mean squared error (RMSE) of 0.06 in the computation of FIT (values range from 0.2-0.6) and a RMSE of 0.54 breaths per minute (bpm) for respiratory rate (values range from 8 - 28 breaths per minute (bpm)) on the MIMIC dataset, and an FIT RMSE of 0.11 and and RR RMSE of 0.66 bpm on the CEBS dataset.

eess.SP↗

arXiv2021

Inferring COPD Severity from Tidal Breathing

Objective: To develop an algorithm that can infer the severity level of a COPD patient's airflow limitation from tidal breathing data that is collected by a wearable device. Methods: Data was collected from 25 single visit adult volunteers with a confirmed or suspected diagnosis of chronic obstructive pulmonary disease (COPD). The ground truth airflow limitation severity of each subject was determined by applying the Global Initiative for Chronic Obstructive Lung Disease (GOLD) staging criteria to the subject's spirometry results. Spirometry was performed in a pulmonary function test laboratory under the supervision of trained clinical staff. Separately, the subjects' respiratory signal was measured during quiet breathing, and a classification model was built to infer the subjects' level of airflow limitation from the measured respiratory signal. The classification model was evaluated against the ground truth using leave-one-out testing. Results: Severity of airway obstruction was classified as either mild/moderate or severe/very severe with an accuracy of 96.4%. Conclusion: Tidal breathing parameters that are measured with a wearable device can be used to distinguish between different levels of airflow limitation in COPD patients.

eess.SP↗