MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Accurate Real Time Localization Tracking in A Clinical Environment Using Bluetooth Low Energy and Deep Learning

Z Iqbal*, D Luo , P Henry , S Kazemifar , T Rozario , Y Yan , K Westover , W Lu , D Nguyen , T Long , J Wang , H Choy , S Jiang , Medical Artificial Intelligence and Automation Laboratory, Department of Radiation Oncology, University of Texas Southwestern Medical Center, Dallas, TX, USA 75390

Presentations

(Sunday, 7/29/2018) 3:00 PM - 6:00 PM

Room: Exhibit Hall

Purpose: Wearable sensors provide an avenue for improving clinical workflow by monitoring patient and staff location. Unfortunately, commercial indoor tracking systems are expensive and difficult to integrate with other technologies. We investigate the feasibility of an affordable real time location system (RTLS) using Bluetooth Low Energy technology and deep learning.

Methods: We developed and installed an RTLS system in our radiation oncology clinic. The system was composed of Raspberry Pis (RPI 3 Model B with Raspbian kernel 4.4.38-v7) equipped with Bluetooth capabilities, and these were connected to a MongoDB database. We used a three-layer convolutional neural network (CNN) followed by an artificial neural network (ANN) with dense layers for tracking signals from Bluetooth tags (RadBeacon Dot). The model was trained using data collected from clinical exam rooms. Finally, we compared this model to classical methods such as signal thresholding and triangulation using the precision, recall, Fâ‚?-score and accuracy metrics.

Results: The total cost of the RTLS system hardware was on the order of $10,000 for the entire clinic. The CNN+ANN model outperformed all other methods for determining the position of the Bluetooth tags. The accuracy for the methods were: CNN+ANN (99.9%), Thresholding (86.1%), CNN only (93.7%), and Triangulation (79.8%). The accuracy results for thresholding and triangulation were further improved by adding majority voting (MV) for 30 seconds of signal: Thresholding+MV (94.9%) and Triangulation+MV (89.5%).

Conclusion: We present an affordable and accurate RTLS system for use in a radiation oncology clinic. We utilized deep learning because Bluetooth signals can have noticeable spatial variations. This system has many potential applications, including tracking physician attendance for chart rounds, reducing average waiting time for a patient, and ensuring patient safety. We will investigate these applications in the future.

Keywords

Localization

Taxonomy

Not Applicable / None Entered.

Contact Email