Detection of Freezing of Gait Using Convolutional Neural Networks and Data From Lower Limb Motion Sensors

Shi, Bohan; Tay, E.B.; Au, Wing Lok; Tan, Dawn; Chia, Nicole Shuang Yu; Yen, Shih-Cheng

doi:10.1109/tbme.2022.3140258

Cited by 32 publications

(39 citation statements)

References 69 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The use of deep learning in the medical and healthcare domain has shown great potential for solving a range of problems, such as detecting specific symptoms or abnormalities 63,64 . However, the interpretability of deep learning models remains a significant challenge, and it is often difficult for clinicians to trust the decisions made by a black-box system.…”

Section: Model Interpretation Using Explainable Ai Approachmentioning

confidence: 99%

BGEM™: Assessing Elevated Blood Glucose Levels Using Machine Learning and Wearable Photo plethysmography Sensors

Shi¹,

Dhaliwa²,

soo³

et al. 2023

Preprint

View full text Add to dashboard Cite

Diabetes mellitus (DB) is the most challenging and fastest-growing global public health challenge. An estimated 10.5% of the global adult population suffers from diabetes, and almost half of them are undiagnosed. The growing at-risk population exacerbated the shortage of health resources, with an estimated 10.6% and 6.2% of adults worldwide having impaired glucose tolerance (IGT) and impaired fasting glycemia (IFG), respectively. All the current diabetes screening methods are invasive and opportunistic and must be conducted in a hospital or a laboratory by trained professionals. At-risk subjects might remain undetected for years and miss the precious time window for early intervention in preventing or delaying the onset of diabetes and its complications. This study was conducted at KK Women’s and Children’s Hospital of Singapore, and five hundred participants were recruited (mean age 38.73 ± 10.61 years; mean BMI 24.4 ± 5.1 kg/m2). The blood glucose levels, for most participants, were measured before and after 75g of sugary drink using both the conventional glucometer (Accu-Chek Performa) and the wrist-worn wearable. The results obtained from the glucometer were used as the ground truth measurements. We propose leveraging photoplethysmography (PPG) sensors and machine learning techniques to incorporate this into an affordable wrist-worn wearable device to detect elevated blood glucose levels (⩾ 7.8mmol/L) non-invasively. Multiple machine learning models were trained and assessed with 10-fold cross-validation using subject demographic data and critical features extracted from the PPG measurements as predictors. Support vector machine (SVM) with a radial basis function kernel has the best detection performance with an average accuracy of 84.7%, a sensitivity of 81.05%, a specificity of 88.3%, a precision of 87.51%, a geometric mean of 84.54% and F-score of 84.03%. Hence, PPG measurements can be utilized to identify subjects with elevated blood glucose measurements and assist in the screening of subjects for diabetes risk.

show abstract

Section: Model Interpretation Using Explainable Ai Approachmentioning

confidence: 99%

BGEM™: Assessing Elevated Blood Glucose Levels Using Machine Learning and Wearable Photo plethysmography Sensors

Shi¹,

Dhaliwa²,

soo³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…The use of deep learning in the medical and healthcare domain has shown great potential for solving a range of problems, such as detecting specific symptoms or abnormalities [37,38]. However, the interpretability of deep learning models remains a significant challenge, and it is often difficult for clinicians to trust the decisions made by a black-box system.…”

Section: Related Workmentioning

confidence: 99%

Assessing Elevated Blood Glucose Levels Through Blood Glucose Evaluation and Monitoring Using Machine Learning and Wearable Photoplethysmography Sensors: Algorithm Development and Validation

Shi¹

2023

Preprint

View full text Add to dashboard Cite

Diabetes mellitus (DB) is the most challenging and fastest-growing global public health challenge. An estimated 10.5% of the global adult population has been suffering from diabetes, and almost half of them are undiagnosed. The growing at-risk population further exacerbated the scarce health resources where the adults worldwide with impaired glucose tolerance (IGT) and impaired fasting glycaemia (IFG) were estimated at around 10.6% and 6.2%, respectively. All the current diabetes screening methods are invasive and opportunistic and must be conducted in a hospital or a laboratory by trained professionals. At-risk subjects might remain undetected for years and miss the precious time window for early intervention in preventing or delaying the onset of diabetes and its complications. This study was conducted at KK Women's and Children's Hospital of Singapore, and five hundred participants were recruited (mean age 38.73 $\pm$ 10.61 years; mean BMI 24.4 $\pm$ 5.1 kg/m\textsuperscript{2}). The blood glucose levels, for most participants, were measured before and after 75g of sugary drink using both the conventional glucometer (Accu-Chek Performa) and the wrist-worn wearable. The results obtained from the glucometer were used as the ground truth measurements. We propose leveraging photoplethysmography (PPG) sensors and machine learning techniques to incorporate this into an affordable wrist-worn wearable device to detect elevated blood glucose levels ($\geqslant 7.8 mmol/L $) non-invasively. Multiple machine learning models were trained and assessed with 10-fold cross-validation using subject demographic data and critical features extracted from the PPG measurements as predictors. Support vector machine (SVM) with a radial basis function kernel has the best detection performance with an average accuracy of 84.7%, a sensitivity of 81.05%, a specificity of 88.3%, a precision of 87.51%, a geometric mean of 84.54% and F-score of 84.03%. Hence, PPG measurements can be utilized to identify subjects with elevated blood glucose measurements and assist in the screening of subjects for diabetes risk.

show abstract

“…Band-pass filtering is often performed on time series data to eliminate low-frequency and high-frequency noise to reduce external influences. However, DL models usually require minimal filtering or introduce noise into the input data, thereby preventing model overfitting and improving generalization performance and robustness 29,30 . Therefore, the data was not filtered at all in this work.…”

Section: Data Screening and Filteringmentioning

confidence: 99%

“…The CNN was widely used in classification, target detection, image segmentation, and other fields due to its advantages of automatically learning complex features. However, training a deeper model requires a large number of computing resources and data 29 . Considering the limitation of the input windows sizes, this work constructed a 6-channel CNN architecture with only 6 convolutional layers according to the data characteristics to avoid overfitting, in which the channel represents the modulus obtained by CWT on the time-series component of the sensor data.…”

Section: Cnn Architecture and Training Strategymentioning

confidence: 99%

An Interpretable Deep Learning Optimized Wearable Daily Monitoring System for Parkinson's Disease Patients

Chen

Sun

et al. 2023

Preprint

View full text Add to dashboard Cite

Walking monitoring in the daily life of patients with Parkinson's disease (PD) is of great significance for tracking the progress of the disease. This study aims to implement an accurate, objective, and long-term walking monitoring algorithm optimized by the interpretable deep learning architecture, and explore the most representative spatiotemporal motor features. Five inertial measurement units attached to the wrist, ankle, and waist are used to collect motion data from 100 subjects during a 10-meter walking test. The raw data of each sensor are subjected to the continuous wavelet transform before training in the constructed deep learning architecture based on a 6-channel convolutional neural network (CNN). The results show that the sensor located at the waist has the best classification performance with an accuracy of 98.01% ± 0.85% and the area under the receiver operating characteristic curve (AUC) of 0.9981 ± 0.0017 under ten-fold cross-validation. The gradient-weighted class activation mapping shows that the feature points with greater contribution to PD were concentrated in the lower frequency band (0.5~3Hz) compared with healthy controls. The visual maps of 3D CNN show that only three out of six time series have a greater contribution, which is used as a basis to further optimize the model input, and the raw data processing cost is reduced by 50% while the classification performance (AUC = 0.9929 ± 0.0019) is still maintained. As far as we know, this is the first study that considers visual interpretation-based optimization of an intelligent classification model in the objective diagnosis of PD.

show abstract

Detection of Freezing of Gait Using Convolutional Neural Networks and Data From Lower Limb Motion Sensors

Cited by 32 publications

References 69 publications

BGEM™: Assessing Elevated Blood Glucose Levels Using Machine Learning and Wearable Photo plethysmography Sensors

BGEM™: Assessing Elevated Blood Glucose Levels Using Machine Learning and Wearable Photo plethysmography Sensors

Assessing Elevated Blood Glucose Levels Through Blood Glucose Evaluation and Monitoring Using Machine Learning and Wearable Photoplethysmography Sensors: Algorithm Development and Validation

An Interpretable Deep Learning Optimized Wearable Daily Monitoring System for Parkinson's Disease Patients

Contact Info

Product

Resources

About