Orthotic Prescription for Pediatric Flexible Flat Feet using Convolutional Neural Networks.

Pediatric flexible flat foot (PFFF) is known to in-crease the foot structure's load, causing potential disability. Foot orthoses are one of the most common non-surgical methods to improve the medial longitudinal arch of the foot for improving PFFF. However, orthoses are not routinely prescribed due to their high cost, and discomfort caused by a restriction of foot movement. Furthermore, there are no quantitative standards or guidelines for an orthotic prescription, which makes the decision-making process of less experienced podiatrists chal-lenging. In this study, the authors investigated convolutional neural networks to classify the needs of orthotic prescription. Using image augmentation techniques and training a VGG-16 model, we achieved high precision and recall, 1 and 0.969 accordingly, to classify orthotic prescription needs.

Depression Level Prediction in People with Parkinson's Disease during the COVID-19 Pandemic.

Many recent studies show that the COVID-19 pandemic has been severely affecting the mental wellness of people with Parkinson's disease. In this study, we propose a machine learning-based approach to predict the level of anxiety and depression among participants with Parkinson's disease using surveys conducted before and during the pandemic in order to provide timely intervention. The proposed method successfully predicts one's depression level using automated machine learning with a root mean square error (RMSE) of 2.841. In addition, we performed model importance and feature importance analysis to reduce the number of features from 5,308 to 4 for maximizing the survey completion rate while minimizing the RMSE and computational complexity.

Machine Learning-based Meal Detection Using Continuous Glucose Monitoring on Healthy Participants: An Objective Measure of Participant Compliance to Protocol.

Meal timing affects metabolic responses to diet, but participant compliance in time-restricted feeding and other diet studies is challenging to monitor and is a major concern for research rigor and reproducibility. To facilitate automated validation of participant self-reports of meal timing, the present study focuses on the creation of a meal detection algorithm using continuous glucose monitoring (CGM), physiological monitors and machine learning. While most CGM-related studies focus on participants who are diabetic, this study is the first to apply machine learning to meal detection using CGM in metabolically healthy adults. Furthermore, the results demonstrate a high area under the receiver operating characteristic curve (AUC-ROC) and precision-recall curve (AUC-PR). A cold-start simulation using a random forest algorithm yields .891 and .803 for AUC-ROC and AUC-PR respectively on 110-minutes data, and a non-cold start simulation using a gradient boosted tree model yields over .996 (AUC-ROC) and .964 (AUC-PR). Here it is demonstrated that CGM and physiological monitoring data is a viable tool for practitioners and scientists to objectively validate self-reports of meal consumption in healthy participants.

Sensor Selection for Activity Classification at Smart Home Environments.

As the world's older population grows dramatically, the needs of continuing care retirement communities increases. Studies show that privacy can be a major concern for adopting technologies, while the older population prefers smart homes [1]. In order to minimize the number of sensors to be installed in each house, we performed Principal Component Analysis (PCA) to filter out the relatively unimportant sensors. We applied a machine learning model to classify residents' activity types, using a different set of sensors chosen by PCA. Then, we validated the trade-off between the classification model accuracy and the number of sensors used in classification. Our experiment shows that feature engineering helps reduce accuracy degradation for activity type classification when using fewer sensors in smart homes.

A Machine Learning Approach to Detecting Low Medication State with Wearable Technologies.

Medication adherence is a critical component and implicit assumption of the patient life cycle that is often violated, incurring financial and medical costs to both patients and the medical system at large. As obstacles to medication adherence are complex and varied, approaches to overcome them must themselves be multifaceted.This paper demonstrates one such approach using sensor data recorded by an Apple Watch to detect low counts of pill medication in standard prescription bottles. We use distributed computing on a cloud-based platform to efficiently process large volumes of high-frequency data and train a Gradient Boosted Tree machine learning model. Our final model yielded average cross-validated accuracy and F1 scores of 80.27% and 80.22%, respectively.We conclude this paper with two use cases in which wearable devices such as the Apple Watch can contribute to efforts to improve patient medication adherence.

A Scalable Smartwatch-Based Medication Intake Detection System Using Distributed Machine Learning.

Poor Medication adherence causes significant economic impact resulting in hospital readmission, hospital visits and other healthcare costs. The authors developed a smartwatch application and a cloud based data pipeline for developing a user-friendly medication intake monitoring system that can contribute to improving medication adherence. The developed Android smartwatch application collects activity sensor data using accelerometer and gyroscope. The cloud-based data pipeline includes distributed data storage, distributed database management system and distributed computing frameworks in order to build a machine learning model which identifies activity types using sensor data. With the proposed sensor data extraction, preprocessing and machine learning algorithms, this study successfully achieved a high F1 score of 0.977 with 13.313 seconds of training time and 0.139 seconds for testing.

Medhere: A Smartwatch-based Medication Adherence Monitoring System using Machine Learning and Distributed Computing.

Poor medication adherence threatens an individual's health and is responsible for substantial medical costs in the United States annually. In order to improve medication adherence rates and provide timely reminders, we developed a smartwatch application that collects data from embedded inertial sensors, which include an accelerometer and gyroscope, to monitor a series of actions happening during an individual's medication intake. After the collected data was delivered to a server, Apache Spark was used to distribute the data and apply machine learning algorithms in order to predict several discrete actions including medication intake. By utilizing these tools, we were able to preprocess high frequency sensor data and apply a random forest algorithm, yielding high frequency and recall of the aforementioned actions.