Spatiotemporal Representations of Rapid Visual Target Detection: A Single-Trial EEG Classification Algorithm

Abstract

Brain computer interface applications, developed for both healthy and clinical populations, critically depend on decoding brain activity in single trials. The goal of the present study was to detect distinctive spatiotemporal brain patterns within a set of event related responses. We introduce a novel classification algorithm, the spatially weighted FLD-PCA (SWFP), which is based on a two-step linear classification of event-related responses, using fisher linear discriminant (FLD) classifier and principal component analysis (PCA) for dimensionality reduction. As a benchmark algorithm, we consider the hierarchical discriminant component Analysis (HDCA), introduced by Parra, et al. 2007. We also consider a modified version of the HDCA, namely the hierarchical discriminant principal component analysis algorithm (HDPCA). We compare single-trial classification accuracies of all the three algorithms, each applied to detect target images within a rapid serial visual presentation (RSVP, 10 Hz) of images from five different object categories, based on single-trial brain responses. We find a systematic superiority of our classification algorithm in the tested paradigm. Additionally, HDPCA significantly increases classification accuracies compared to the HDCA. Finally, we show that presenting several repetitions of the same image exemplars improve accuracy, and thus may be important in cases where high accuracy is crucial.

Multimodal Neural Network for Rapid Serial Visual Presentation Brain Computer Interface

“…One such passive BCI application is the Rapid Serial Visual Presentation (RSVP) application (Parra et al, 2007; Bigdely-Shamlo et al, 2008; Alpert et al, 2014). In RSVP tasks, a subject is instructed to search and count for target images within a continuous stream of images, displayed at a fast pace, e.g., at 10 Hz.…”

“…The framework showed success in RSVP tasks for triaging image databases of natural scenes (Gerson et al, 2006), aerial images (Parra et al, 2007), and missile detection in satellite images (Sajda et al, 2010). Alpert et al (2014) presented a a two step linear classification algorithm for RSVP tasks. The Spatially Weighted FLD-PCA (SWFP) algorithm first learns a spatio-temporal weights matrix that amplifies important locations for classification in both spatial and temporal domains.…”

“…The network uses a spatio-temporal regularizer for EEG that reduces overfitting and increases performance. The network was compared to Alpert et al (2014) using the same data set, showing improved performance.…”

A review of feature extraction and classification algorithms for image RSVP-based BCI

“…Hierarchical discriminant component analysis has been proposed in [40], where this method estimates EEG signatures of target detection events using multiple linear discriminators, each trained at a different time window relative to the image onset. Since EEG signals contain both spatial and temporal information, a spatiotemporal representation for RSVP-based EEG data has been proposed by Alpert [37]. This representation is divided into two steps: (1) LDA is applied at each timestamp to produce the spatial weights and a spatial weight matrix is then used for mapping original epoch to a new space, and (2) PCA is then used for dimensionality reduction based on the temporal domain, i.e., for each independent channel.…”

“…PCA has been applied in EEG signal analysis for dimensionality reduction [35] and the production of spatial filters [36]. In RSVPbased BCI literature, PCA has only been applied for feature dimension reduction to date [1,37].…”