Automatic apraxia detection using deep convolutional neural networks and similarity methods

Dementia represents one of the great problems to be solved in medicine for a society that is becoming increasingly long-lived. One of the main causes of dementia is Alzheimer's disease, which accounts for 80% of cases. There is currently no cure for this disease, although there are treatments to try to alleviate its effects, which is why detecting Alzheimer's disease in its early stages is crucial to slow down its evolution and thus help sufferers. One of the symptoms of the disease that manifests in its early stages is apraxia, difficulties in carrying out voluntary movements. In the clinical setting, apraxia is typically assessed by asking the patient to imitate hand gestures that are performed by the examiner. To automate this test, this paper proposes a system that, based on a video of the patient making the gesture, evaluates its execution. This evaluation is done in two steps, first extracting the skeleton of the hands and then using a similarity function to obtain an objective score of the execution of the gesture. The results obtained in an experiment with several patients performing different gestures are shown, showing the effectiveness of the proposed method. The system is intended to serve as a diagnostic tool, enabling medical experts to detect possible mobility impairments in patients that may have signs of Alzheimer's disease.