Researchers from the Epilepsy Centre at Kuopio University Hospital, University of Eastern Finland and Neuro Event Labs created an algorithm with deep convolutional neural networks to offer fast, reliable and automatic assessment of the severity of myoclonic jerks. The neural network and pre-trained models could identify and track key points in the human body.
Predefined keypoints and raw data and algorithmic output of wrist movement. Source: Neuro Event Labs
Myoclonus is brief, involuntary twitching of muscles. This is the most disabling and progressive drug-resistant symptom in patients with progressive myoclonus epilepsy type 1 (EPM1). Myoclonus is stimulus sensitive and its severity fluctuates throughout the day. Stress, sleep deprivation and anxiety can aggravate the symptoms.
The clinical analysis of myoclonus is challenging and requires extensive expertise from doctors. The medical community has been searching for an automatic tool to improve the consistency and reliability of serial myoclonus evaluations. Currently, the unified myoclonus rating scale (UMRS) is the gold standard for evaluating myoclonus. UMRS is a clinical video recorded test panel.
To test the algorithm, the team analyzed 10 video recorded UMRS test panels using automatic post estimation and key point detection methods. The automatic methods were successful at detecting and tracking predefined key points in the human body while moving. It also successfully analyzed speed changes and smoothness of the myoclonic movements during an active seizure. Scores were gathered with automatic myoclonus detection that correlated with clinical UMRS myoclonus test scores evaluated by a clinical researcher.
The study proved that an automatic detection method with keypoint detection and pose estimation from video footage was able to quantify myoclonic jerks in EPM1 patients.
A paper on this algorithm was published in Seizure: European Journal of Epilepsy.
