Project Objectives:

The purpose of the study is to determine if a suite of few sensors embedded in a 3D printed medication bottle can accurately determine the time and capture the motion associated with a medication-taking person. The circuit that combines all sensors is small in size 3x3 inches. The system sends acceleration, weight, and open/close measurements several times a second to a nearby computer. A goal of the study is to understand the performance of the sensors or any combination of them as a building block for medication taking activity monitoring.

Technology Rationale:

Recent years have seen the size, cost and energy consumption of small wireless sensor decrease by several orders of magnitude. Indeed, today, wireless sensors can be bought for an affordable price. These improvements have made it possible to connect everyday objects to the Internet, resulting in the visionary concept of the Internet of Things (IoT). While there are many possible uses of the IoT for art, education, and security, measuring everyday activity to improve the health and wellbeing of persons is rapidly becoming an active area of research. Medication adherence is an important component of health and wellbeing, with voluminous studies showing the important of adequate medication adherence. Current quantification of medication adherence is enormously obstructive and resource-consuming, were patients must use wearable sensors continuously. This work seeks to quantify the ability of few styles of IoT sensors to enable accurate measurements and monitoring of medication taking activity at a much lower cost than was previously possible and without invasiveness. If medication adherence can be determined using IoT style sensing, new forms of self-management and clinical care would be enabled. We built a pill bottle using a 3D printer and equipped it with a combination of digital and analog sensors for medication adherence monitoring applications. We aim at developing a low-energy sensor system based on the PIP tag as the platform.



Technical Approach:
We aim at using the data from the wireless sensors embedded in the 3D printed pill bottle for the following:

1. An off-body Sensing-Based Framework for medication compliance monitoring: Studying the hand gestures associated with medication taking. From these gestures we try to recognize the actions of twisting the bottle cap, pouring pill into hand and hand to mouth. In this regard, we propose a compliance sensing framework that does not require on-body sensing modalities or costly installation and maintenance. Further, we address the following questions

  • How framework performance affected as we use a binary classifier?
  • How framework performance affected as we use other classifiers, for multi-class classification?

2. PatientSense: Patient Discrimination from In-Bottle Sensors Data. We aim at using the data obtained from the system, especially the acceleration data, for user recognition and authentication (who took the medicine and when?)

We investigate identification of persons taking medication using a sensor-equipped pill bottle. To evaluate our system, we address the following four questions:

  • How is accuracy affected as the number of users increases?
  • How is system performance impacted with respect to the training size?
  • How is the system performance impacted with respect to the number of accelerometers used, i.e. one vs. two?

Results To Date and Future Work Plan:

This project has led to one paper in DFHS'19 (a workshop co-located with BuildSys'19). We report the performance and our framework using our smart pill-bottle prototype. Figure 1 depicts the framework, Figure 2 shows the classification accuracy using one accelerometer and two accelerometers. Figure 3 shows the compliance vs. non-compliance recognition accuracy by using multi-class classifiers.

The project also led to another paper in MobiQuitous'19. In Figure 4, we show the overview of PatientSense. In Figure 5, we show the user discrimination performance of PatientSense. In addition, our system achieves high accuracy with different training percentages as shown in Figure 6. Finally, Figure 7 shows the performance of our system by using individual sensor.

We will use a round bottle that we recently built to measure the performance of our approach over different days.

1. A Framework for Medication-Compliance Monitoring

2. PatientSense