Accurate real time localization tracking in a clinical environment using Bluetooth Low Energy and deep learning

Area of Science:

• Medical technology
• Artificial intelligence in healthcare
• Radiation oncology informatics

Background:

• Deep learning applications are expanding in medicine.
• Accurate tracking of patients and staff is crucial for workflow and safety in clinical settings.
• Bluetooth Low Energy (BLE) tags offer a potential solution for real-time location systems (RTLS).

Purpose of the Study:

• To investigate the feasibility of using ANNs and CNNs for tracking BLE tags in a radiation oncology clinic.
• To compare the performance of deep learning models against traditional methods like RSSI thresholding and triangulation.
• To develop an accurate and efficient RTLS for improving healthcare operations.

Main Methods:

• Utilized artificial neural networks (ANNs) and convolutional neural networks (CNNs) for BLE tag localization.
• Compared a combined CNN+ANN model with individual CNN, RSSI thresholding, and triangulation methods.
• Incorporated temporal information into the deep learning models for enhanced accuracy.

Main Results:

• The combined CNN+ANN network achieved a 99.9% accuracy in identifying BLE tag locations.
• Deep learning models significantly outperformed traditional methods, including CNN (94%), thresholding (95%), and triangulation (95%) with majority voting.
• The system demonstrated high precision by utilizing temporal data.

Conclusions:

• A deep learning approach, specifically a combined CNN+ANN model, is highly effective for accurate real-time patient and staff tracking using BLE tags.
• This technology has the potential to significantly improve clinical workflow, operational efficiency, and patient safety in radiation oncology.
• The developed RTLS is affordable and suitable for future deployment in hospital settings.