SCH: MS-ADAPT: Multi-Sensor Adaptive Data Analytics for Physical Therapy

The current project addresses the needs of the US population that suffers from low back pain (LBP). Chronic recurrent LBP is a significant public health problem that results in functional limitations and disability, as well as financial burden for the individual and society. Up to 80% of people will experience LBP at some point in their lifetime, and the total costs of LBP in the U.S. exceed 100 billion dollars per year, including lost wages resulting from an inability to work.

Physical therapy is effective for managing chronic LBP and improving patient outcomes. However, patient adherence to physical therapist (PT) recommendations is low, and research suggests that real-time personalized feedback on posture and movement throughout the day can improve outcomes. However, there are major limitations of existing sensor systems.

Additionally, there is a need to integrate these sensor data with existing sensor technologies and clinical measures. To address this need, this project will design and develop a sensor system that supports remote monitoring of posture and movement in patients with LBP, patient adherence to PT recommendations, and the impact of adherence on outcomes.

This system will both advance the science through development of non-invasive, low-profile sensors to measure low back posture and movement in a real-life setting, and connect this novel sensor information with existing devices and clinical measures that are used to monitor activity and the impact of pain in patients with LBP. Further, new diagnostic and treatment information gathered from these novel sensors could be used to advance treatments and improve outcomes for people with LBP.

This research could also be extended to management of other serious health conditions, such as amputations, spinal cord injury, and stroke. This work will result in innovations in wearable technologies, deep learning, system integration, and human-computer interfaces. Notably, we will validate the fabric sensors for distributed motion monitoring using Electrical Impedance Tomography and develop algorithms to capture not how much strain the wearer is experiencing but also the direction of these strains.

For predictive modeling, we will use mathematics and a statistical approach that maps the received data to the category of strain being experienced. In addition to advancing the science and clinical practice, this project will contribute to the training of a new generation of interdisciplinary researchers at the intersection of engineering and health sciences across PhD, Masters, and undergraduate students, and health professional trainees.

The MS-ADAPT system is proposed as a human-in-the-loop cyber-physical system that integrates data from novel fabric sensors with data from wrist accelerometers and app-based patient-reported outcomes, and uses machine-learning analytics to enable predictions in support of personalized physical therapy. The work is structured into five research aims: 1) fabric sensors for functional movement assessment of the lumbar spine: distributed sensing is achieved by forming a network of smart “K-Tape sensors”, which are strain-sensitive nanocomposites integrated with commercial kinesiology tape for characterizing skin strains, movements, and muscle activity, 2) data integration and visualization: a platform for integrating multi-model data (changes in skin resistance and strain maps from smart K-Tape sensors; accelerometry from Fitbit; and patient-reported outcomes from apps) and PT visualizations to support decision-making; 3) laboratory assessment to interpret smart K-Tape data for objective assessment of lumbar spine movement and muscle activity, 4) novel machine-learning models that reflect lumbar spine biomechanics: create both physics-based models and deep learning models to predict posture and movement type, muscle activation, movement magnitude, and quality, and 5) clinical evaluation: assess posture and movement in a free-living environment, adherence to PT recommendations, and the association with improvement in terms of LBP symptoms and function.

Many technical challenges exist with respect to characterizing new sensors and using novel data streams for precision medicine insights. Both physics-based models and CNN-LSTM models will be created and compared for predicting the movement type and quality and assessing adherence to PT recommendations throughout the day.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.