Personalized Treatment through Biosensors and Machine Learning ML

Abstract

Precision medicine and personalized treatment can be effectively achieved through a combination of machine learning ML techniques and data gathered through specialized biosensors. However, the biggest challenge appears to be need of better data – in terms of both quality and quantity – on which to apply appropriate ML techniques. While ML can uncover deeper dependencies between the data, it requires enough good data to capture these dependencies. No matter how good the ML algorithms are, they cannot find something which is not in the training set. Hence, better (in terms of both quality and quantity) datasets are required to train algorithms. This, in turn, means that improved biosensors are needed to deliver such comprehensive, specialized data. The first section of the report provides a comprehensive literature review of personalized medicine applications employing ML methods on sensor data, under categories of Mobile sensing and portable devices, Neuroimaging, Brain Machine Interface, Omics and electronic health records, and Biosensor systems. While the first section is focused on opportunities and successes described in the literature, the second section highlights the key challenges which need to be addressed in order to take full advantage of the benefits ML can offer in the areas of personalized medicine. They also underscore the need for biosensors which provide better quality and quantity of data on which ML may operate. The third section describes select “best practices” for ML applications: data quality management and validation, data visualization, and privacy preserving data analysis. Very effective interactive visualization can be delivered through a Shiny framework and R packages, and we present a representative interactive visualizer using real data. Lastly, the appendices provide additional information and references on data quality procedures, data visualization techniques, and topics in ML.