Room: Track 5
Purpose: To present an in-house developed indoor Real-Time Location System (RTLS) to track tagged patients, staff and assets to improve clinic workflow, patient safety and experience .
Methods: Throughout our department building, we installed a grid of Bluetooth Low Energy (BLE) sensors and developed from ground up a location software system that enabled us to track tags transmitting Bluetooth radiofrequency signals in the building. The challenge is that the weak relationship between signal strength and distance is significantly altered because of signal reflection and attenuation by building structures. It is very difficult if not impossible to locate a tag using conventional distance-based deterministic or triangulation algorithms as GPS does. However, we noticed that received signal strengths by sensors of a tag at a given zone form a highly distinguishable pattern. So we developed an Artificial Intelligence (AI)-based location engine to drive all related clinical applications using RTLS information. The three-floor clinic building was equipped with 142 BLE sensors and divided into 114 functional zones where signal strengths from all sensors were collected. We trained a Deep Neural Network (DNN) consisted of an embedded Long Short-Term memory (LSTM) unit and a deep classifier using these signal data as inputs and corresponding zones as outputs. The inputs were augmented to generate cross-zone sequences to mimic real-word cases. We further constrain the results using the spatial connection information to increase stability. The network is then used to calculate the real-time location.
Results: The RTLS engine was able to identify tag’s location. The system achieved an accuracy of 100% when the lagging is less of a concern, compared with about 95% for a baseline CNN+ANN model.
Conclusion: We established a Real-time Location System in a radiation oncology clinic building and has been regularly used in clinical practices.