Sounds of seizures-acoustic information enables immediate recognition and detection of generalized tonic-clonic seizures in dogs

Epilepsy is the most common chronic neurological disease in humans
and dogs. Seizures greatly impact patients' and caretakers' quality of
life. One of the caretakers' major concerns is that seizures may remain
unnoticed or may be noticed too late. There is need for reliable detection
of seizures in order to apply quick emergency treatment for
preventing further seizure evolution resulting in cluster seizures or
status epilepticus. Artificial intelligence may represent a new tool to
overcome this gap. A crucial step in this direction is the creation of
quality datasets and computational tools for investigating seizurerelated
audio, video, and sensor signals. A dataset of 42 audio tracks
of videos of dogs with generalized tonic-clonic epileptic seizures was
collected and annotated. 138 statistical features were used and
9 classifier types using 4474 sound samples with the duration of
1 second each were investigated. The obtained classifiers were evaluated
with k-fold cross-validation and automatic hyperparameter tuning,
reaching balanced accuracy of above 70%. Classical machine
learning methods show promising results in the detection of epileptic
seizure sounds, providing a potential basis for the development of an
alert and detection system for domestic and clinical environments. In
addition, such systems could significantly improve the detection of
seizure frequency and semiology to measure therapeutic effectiveness
in more scientific settings. Employing various advanced deep
learning techniques and neural networks has even greater potential
to increase accuracy; however, larger datasets are needed.