Development of a Quantitative Tool Based on Deep Learning for Automatic Apraxia Detection (DLAAD)

Dementia is a priority in medicine today, with Alzheimer's Disease (AD) being the main cause, accounting for 80% of cases. One of the symptoms of this disease, from its early stages, is apraxia, which produces a difficulty in performing voluntary gestures, which a person without the disease could perform without difficulty. This is why neurologists try to detect possible apraxias early to try to diagnose the disease as soon as possible. For this, the patient is asked to imitate a gesture, usually given through a picture, the execution of which is assessed by a neurologist. However, there are limitations such as the difficulty in assessing doubtful cases or comparing serial tests, since the subjectivity of the evaluator comes into play. We present a system based on convolutional neural networks to make an automatic evaluation of the gestures made by patients, captured by a mobile phone camera. The system takes into account both the movements and the time spent in order to give an assessment. The results obtained are promising, reaching a 73% hit rate for a particular gesture. Finally, we propose a way to improve the system, which will be worked on in the near future.