Analyze-Me
Software tools for processing physiological signals from patients with a non-verbal form of autism
Phase 2
Type of project: Prototype
Disability concerned: Autism and Pervasive Developmental Disorders, Mental disability, Polydisability
Topics: Autonomy, Services and communication, Health, Interpersonal relationships
Status: Completed
Our project aims to develop software for the detection and visualisation of stress induction or attempted communication. By extracting the characteristics of physiological signals, our tool provides them as input to a Deep Learning algorithm for classifying the communication vs. difficult behaviour of non-verbal autistic patients.
Patients with non-verbal autism often have difficulty communicating, which can lead to the development of challenging behaviours that can be a danger not only to the patient but also to their carers. To address this problem, we have developed a graphical user interface that enables real-time data collection from the Empatica bracelet.
This interface processes a variety of physiological signals, including electrodermal activity (EDA), blood volume pulse (BVP), temperature and 3D acceleration (ACC), extracting more than 100 different signal features and allowing visualisation of selected features.
Preliminary data collected from two autistic patients and eight healthy volunteers – an effort approved by the ethics committee – led to the following conclusions:
- The physiological responses of healthy volunteers and autistic patients appear to be significantly different. The mixed data from healthy volunteers and autistic patients did not allow a precise classification of stressful events versus periods of calm.
- However, by analysing the data from each autistic patient individually, we obtained an excellent classification of stressful events in relation to periods of calm, with an accuracy of 98%. In addition, we were able to classify emotional stress and attempts at non-verbal communication with 97% accuracy.
