Objectives/Aims To the best of our knowledge, there are no studies using intelligent sensing to diagnose ADHD. The aim of this interdisciplinary (Medical and Engineering) research is to contribute intelligent sensing based multimodal (audio, video, touch, & text) data and trustworthy and reproducible AI in diagnosing ADHD.

Methods Unmedicated subjects with ADHD were interviewed at the Intelligent Sensing Lab in Newcastle University UK and the multimodal data were captured (figure 1). They completed CANTAB tasks, watched stimulating and neutral videos to elicit hyperfocus and distraction. Other distraction cues like noise, images on a monitor were introduced. Healthy volunteers were used as controls. Data were analysed on speech analysis, action analysis and facial action coding system. We used the existing multimodal signal and information processing algorithms developed at Newcastle University

• Download figure
• Open in new tab
• Download powerpoint

Abstract 24 Figure 1

Recording room layout for the intelligent sensing (audio-video-touch/keypad sensors) ADHD datasets

Results The accuracy of the Audio-Based ADHD Diagnosis reached over 80%. The accuracy of the ADHD Action-Based Diagnosis System reached over 90%. The accuracy of the Facial Action Coding System was 94%.

Conclusions Even with a small number of subject and controls, we were able to develop a proof of the concept system which generates high accuracy results. We are in the process of multimodal fusion and conduct a much larger study.