Early illness recognition using in-home monitoring sensors and multiple instance learning

Abstract

Background: Many older adults in the US prefer to live independently for as long as they are able, despite the onset of conditions such as frailty and dementia. Solutions are needed to enable independent living, while enhancing safety and peace of mind for their families. Elderly patients are particularly at-risk for late assessment of cognitive changes.

Objectives: We predict early signs of illness in older adults by using the data generated by a continuous, unobtrusive nursing home monitoring system.

Methods: We describe the possibility of employing a multiple instance learning (MIL) framework for early illness detection. The MIL framework is suitable for training classifiers when the available data presents temporal or location uncertainties.

Results: We provide experiments on three datasets that prove the utility of the MIL framework. We first tuned our algorithms on a set of 200 normal/abnormal behavior patterns produced by a dedicated simulator. We then conducted two retrospective studies on residents from the Tiger Place aging in place facility, aged over 70, which have been monitored with motion and bed sensors for over two years. The presence or absence of the illness was manually assessed based on the nursing visit reports.

Conclusions: The use of simulated sensor data proved to be very useful for algorithm development and testing. The results obtained using MIL for six Tiger Place residents, an average area under the receiver operator characteristic curve (AROC) of 0.7, are promising. However, more sophisticated MIL classifiers are needed to improve the performance.