A knowledge-based approach for automatic quantification of epileptiform activity in children with electrical status epilepticus during sleep

Abstract

Objective. Electrical status epilepticus during sleep (ESES), as electroencephalographic disturbances, is characterized by strong activation of epileptiform activity in the electroencephalogram during sleep. Quantitative descriptors of such epileptiform activity can support the diagnose and the prognosis of children with ESES. To quantify the epileptiform activity of ESES, a knowledge-based approach to mimic the clinical decision-making process is proposed. Approach. Firstly, a morphological operations-based scheme is designed to quickly locate the positive peaks/negative pits and roughly estimate the onset/offset of spike and slow-wave abnormalities. Then, to provide the accurate duration of ESES patterns, a set of rules for further adjusting these onsets/offsets are proposed by merging medical knowledge with a generalized threshold obtained from statistics. As such, the quantification is accomplished by evaluating the obtained spike and slow-wave abnormalities and their various durations. Main results. The effectiveness and feasibility of the proposed method were evaluated on a clinical dataset that collected at Children’s Hospital of Fudan University, Shanghai, China. We demonstrate that the proposed method can recognize different types of spike and slow-wave abnormalities. The sensitivity, precision, and false positive rate achieved 91.96%, 97.09%, and 1.88 min−1, respectively. The estimation error for the spike-wave index was 2.32%. Comparison results showed that our method outperforms the state-of-the-art. Significance. The quantification of spike and slow-waves provides information about ESES activity. The detection of variations types of spike and slow-waves improves the performance in the quantification of ESES. Experimental results suggest that the proposed method has great potential in automatic ESES quantification and can help improve the diagnosis and researches of epileptic encephalopathy with ESES.

A Hybrid Expert System for Individualized Quantification of Electrical Status Epilepticus During Sleep Using Biogeography-Based Optimization

“…A. EEG quantification scheme using expert knowledge and morphology analysis A brief description of MA-EDM is presented below, more details can be found in our previous work [21].…”

“…For CMM-BBO, several reasonable values, which are suggested in the application of BBO algorithm [39], are adopted to set the following control parameters for the current implementation: population size N P = 100, mutation probability mmax = 0.005, emigration rate E and immigration rate I for habitat = 1, the control parameter of CMM-based migration operator Pe = 0.5, elitism parameter p = 1, and maximum number of objective function evaluations = 15000. For the morphology analysis-based expert decision model, the bidirectional amplitude threshold T hu = 71 and T h l = -25 is taken, which follows the convention employed for ESES analysis [21]. The presented hybrid expert system is implemented and evaluated in MATLAB R2019a.…”

“…Manually predefined values of lengths may not obtain effective features of ESES patterns, and the height of the structuring element used could also affect system performance. The influence of using different heights for peak/pit detector had been explored and shown in our previous work [21]. Moreover, the same and empirical horizon is used by the system MA-EDM for all features, which overlooks the differences between peaks and pits as well as the variation between individuals.…”

“…As ESES is one of the rare epilepsies, the related and welllabeled data are challenging and expensive to obtain, in which case a suitable morphology analysis-based expert decision model (MA-EDM) can be a better solution to identify and quantify ESES patterns [21]. Results from our previous work have demonstrated their great potential for the application of the quantification of ESES patterns.…”

Untitled

“…59 Semiautomated spike detection software was also recently developed to compare with human readers with no significant differences in the calculated SWI, suggesting an automated approach may be used in the future to more consistently and accurate calculate the SWI. 61,62 Fig. 3 A 4-year-old girl child patient presented at Nationwide Children's Hospital with subacute onset of "verbal auditory agnosia" (difficulty understanding spoken language), behavioral deterioration, impulsivity, and hyperactivity.…”