Selective sensing of a heterogeneous population of units with dynamic health conditions

Monitoring a large number of units whose health conditions follow complex dynamic evolution is a challenging problem in many healthcare and engineering applications. For instance, a unit may represent a patient in a healthcare application or a machine in a manufacturing process. Challenges mainly arise from: (i) insufficient data collection that results in limited measurements for each unit to build an accurate personalized model in the prognostic modeling stage; and (ii) limited capacity to further collect surveillance measurement of the units in the monitoring stage. In this study, we develop a selective sensing method that integrates prognostic models, collaborative learning, and sensing resource allocation to efficiently and economically monitor a large number of units by exploiting the similarity between them. We showcased the effectiveness of the proposed method using two real-world applications; one on depression monitoring and another with cognitive degradation monitoring for Alzheimer’s disease. Comparing with existing benchmark methods such as the ranking-and-selection method, our fully integrated prognosis-driven selective sensing method enables more accurate and faster identification of high-risk individuals.