Abstracts

To evaluate the feasibility of recognizing seizures based on quantitative analysis of digital video recordings, and differentiating them from voluntary movements., We analyzed digital video recordings of 8 seizures (confirmed by simultaneous EEG) from six patients, and 8 voluntary movements from two control subjects. Motion was quantified by extracting temporal motion strength (MS) and motion trajectory (MT) signals. MS signals that measure the area of moving body parts were obtained by parametric motion segmentation based on pre-computed optical flow. Parameter vectors obtained from the most reliable among non-overlapping blocks were clustered by the [italic]k[/italic]-means algorithm, and based on the resulting partitions pixels of each frame were assigned to the moving body parts or to the background. MT signals were obtained by selecting anatomical sites on the moving body parts by an automated procedure, and subsequently tracking these sites with trackers capable of adjusting to illumination and contrast changes. For each seizure and voluntary movement (total 16), MS and MT signals were analyzed in the time domain to compute 3 quantitative features: energy ratio (ER), variance of time intervals (VTI) and maximum spike duration (MSD). ER was computed as the ratio of the energy contained in the last 75% of the samples of the autocorrelation sequence to the energy contained in the first 25% of samples; ER is expected to differentiate clonic movements (large values due to rhythmicity) from isolated jerky movements (small values). VTI was obtained from the length of the time intervals between adjacent spikes (MS signals) or adjacent extrema (MT signals); VTI measures rhythmicity in movements since they produce values close to zero. MSD provides a quantitative measure of the movement speed, differentiating rapid movements of shorter duration (small values) from slower movements of longer duration (large values)., Class separation was visualized by plotting the quantitative features derived from MS signals (ER[sub]ms[/sub]-VTI[sub]ms[/sub], ER[sub]ms[/sub]-MSD[sub]ms[/sub]) and MT signals (ER[sub]mt[/sub]-VTI[sub]mt[/sub], ER[sub]mt[/sub]-MSD[sub]mt[/sub]) in scatter plots (figure). Overall, seizures were linearly separable from voluntary movements. However, one seizure was not clearly separable due to higher VTI[sub]ms[/sub], VTI[sub]mt[/sub], MSD[sub]ms[/sub], and MSD[sub]mt[/sub] values related to the slow, long-duration tonic posturing of the extremities in that seizure., Seizures associated with sudden isolated jerks and tonic-clonic movements can be successfully differentiated from voluntary movements. This study provides groundwork for developing an automated system capable of seizure surveillance and alerting.[figure1],