From data to diagnosis: A comprehensive review of machine learning-driven wearable sensors in healthcare

Comprehensive Summary

There has been a shift towards more aging populations recently where in order to evaluate diseases and clinically actionable information, new health care delivery models need to be incorporated. The conventional healthcare delivery models have limitations and are unable to adequately address personalized solutions as well as early diagnosis and disease management. For these reasons, the authors suggest machine learning algorithms to determine information from intelligent wearable devices that contain sensing technologies. This model is able to provide information from early diagnosis to clinical decision support, which is beneficial for the patient and physician. Research has proven that chronic care management is able to be improved with such a system. This closed-loop system is able to record signals from but not limited to electrocardiograms, electroencephalogram, blood glucose, blood oxygen, and blood pressure where all of the raw data containing noise is able to be filtered out with a machine learning approach using algorithms. The algorithms mentioned include Ensemble Boosted Trees containing multiple decision trees, Artificial Neural Networks that analyze complex signals, Gaussian Process Regression where uncertainty can be converted into a framework, and the LGBM algorithm with efficiency advantages. With the advancement of wearable systems and machine learning algorithms, chronic disease care has been improved due to the evolution of wearable systems from simply monitoring someone's physiological condition to active clinical management. The authors state that these wearable systems, for instance, are currently able to track blood glucose levels in real time as well as noticing fluctuation patterns, predicting future trends, and give management recommendations that are personalized. Although the incorporation of machine learning in wearable systems seems promising, there are significant challenges that still need to be addressed such as the application of the models to new and unseen data to different populations without suffering from performance degradation. Even with the limitations, the authors still argue that with further experimentation and tweaks to the algorithms when applying new information, these models can be beneficial in the medical field with the application of machine learning to wearable systems.

Outcomes and Implications

The advancements of machine learning and wearable sensor technologies in modern healthcare are substantial. These models are applied to a medical setting as they are able to detect physiological conditions as well as personalized management recommendations. The authors emphasize that this can serve as a crucial model in chronic disease management and help healthcare providers come up with personalized solutions. However, the limitations need to be taken into account where the models cannot be generalized to populations where there has been unseen data. Through further experimentation and development of these models, individuals should be better able to analyze their chronic illnesses with the help of healthcare providers.