Supporting Physical Exercise in Patients with Neuromuscular Diseases: AI-assisted training

/by

Berardina De Carolis1, Giuseppe Palestra1, Erika Schirinzi2, Gabriele Siciliano2, Mario A. Bochicchio1
1. Department of Computer Science, University of Bari Aldo Moro, Italy
2. Department of Clinical and Experimental Medicine, Neurological Clinic, University of Pisa

In Neuromuscular Disorders (NMDs) physiotherapy plays a crucial role in preserving patients’ mobility
and mitigating the adverse impacts of immobility and contractures.
Regular stretching and, in select cases, low-impact exercises are recommended to slow disease progression.

Investigate on the potential of digital technologies to alleviate the time burden on both patients and
healthcare providers
Understand whether AI-based solution can facilitate remote patient monitoring and enhance the frequency
of care, particularly for those facing challenges with travel logistics.
AI-Assisted Training
AI system tailored to assist patients and their
caregivers in performing physical exercises.
• Real-time video feed analysis of the patient’s exercise
routine
• Assess in real-time of the accuracy of the patient’s
movements
• Guidance and corrections when errors (angles) are
detected.
• Tracking and log of eight key parameters (i.e., angles
between the torso and limbs), which can aid in
monitoring the progressive decline in muscle function
over time.
The system is GDPR-compliant and is poised for
pre-clinical evaluation, marking a promising step
towards more effective and accessible NMD care
management.
The Dataset
In this stage of the project we collected data from
healthy subjects doing fitness exercises.
• 10 Exercises
• 1.132 videos recorded
• Data Augmentation -> 13.584 videos
• Video split in 15-frames windows
• 548.954 windows in total
• Window balancing -> 38.000 windows
x exercise
• Windows Shape (15,51+16)
Comparing Models
Future Work
– Training on exercises typical of NMD intervention
– Integration with health monitoring devices
(smartwatch)
– Controlled study
– Integration in Telemedicine applications
Model Accuracy Precision Recall F1
Combo LSTM 0.953 0.961 0.954 0.955
GRU 0.940 0.947 0.941 0.942
LSTM 0.833 0.842 0.835 0.835