Automatic Segmentation of Episodes Containing Epileptic Clonic Seizures in Video Sequences

Automatic Segmentation of Episodes Containing Epileptic Clonic Seizures in Video Sequences 150 150 IEEE Transactions on Biomedical Engineering (TBME)


Kalitzin,S.; Petkov,G.; Velis,D.; Vledder,B.; Lopes da Silva,F.
Volume: 59, Issue: 12, Pages: 3379-3385,  Abstract  
Publication Year: 2012

Epilepsy is a debilitating clinical condition of the central nervous system characterized by unexpected sudden display of fits, known as epileptic seizures, which partially or entirely impair normal functioning. Motor seizures, characterized by uncontrollable and sometimes prolonged violent body movements,  are those that have perhaps the most dramatic appearance and may pose a hazard to the individual’s health. In many cases trained personnel or family members have to be involved as caregivers.  One possibility to improve the quality of life of both patients and their caregivers is to develop  intelligent automated seizure-alert systems that are capable of processing in real time data acquired from sensors with preferably no direct physical contact with the patient.

We present here a novel remote-sensing paradigm for detection of a particular class of motor seizures, called clonic seizures,  characterized by rhythmic paroxysmal motor activity.  Live video signals are acquired by commercially available video equipment and optical flow analysis estimates the velocities of moving objects captured in the video sequences. Unique component in our approach is the  extraction of the rates of global features of motion such as rotation, translation, dilatation, and shear.  By further processing the data with non-stationary wavelet techniques (shown in the upper frame of the figure  that corresponds to analysis of a video sequence from which  three selected frames of a major motor seizure  are shown in the lower frame), we are able to reliably identify those fragments that may contain the presence of movements characteristic of clonic seizures.

Our algorithms are designed for feed-forward processing allowing their implementation in real-time seizure-alerting systems. In off-line applications the same system can be employed for automated screening of large amounts of video sequences taken for diagnostic purposes in clinical facilities, thereby increasing the efficiency of laborious and time-consuming diagnostic routines in a hospital environment.