RESUMO
Blepharospasm (BSP) is an adult-onset focal dystonia with phenomenologically heterogeneous effects, including, but not limited to, blinks, brief or prolonged spasms, and a narrowing or closure of the eyelids. In spite of the clear and well-known symptomatology, objectively rating the severity of this dystonia is a rather complex task since BSP symptoms are so subtle and hardly perceptible that even expert neurologists can rate the gravity of the pathology differently in the same patients. Software tools have been developed to help clinicians in the rating procedure. Currently, a computerised video-based system is available that is capable of objectively determining the eye closure time, however, it cannot distinguish the typical symptoms of the pathology. In this study, we attempt to take a step forward by proposing a neural network-based software able not only to measure the eye closure, time but also to recognise and count the typical blepharospasm symptoms. The software, after detecting the state of the eyes (open or closed), the movement of specific facial landmarks, and properly implementing artificial neural networks with an optimised topology, can recognise blinking, and brief and prolonged spasms. Comparing the software predictions with the observations of an expert neurologist allowed assessment of the sensitivity and specificity of the proposed software. The levels of sensitivity were high for recognising brief and prolonged spasms but were lower in the case of blinks. The proposed software is an automatic tool capable of making objective 'measurements' of blepharospasm symptoms.
