Switch or stay? Automatic classification of internal mental states in bistable perception

Susmita Sen, Syed Naser Daimi*, Katsumi Watanabe, Kohske Takahashi, Joydeep Bhattacharya, Goutam Saha

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

9 Citations (Scopus)


The human brain goes through numerous cognitive states, most of these being hidden or implicit while performing a task, and understanding them is of great practical importance. However, identifying internal mental states is quite challenging as these states are difficult to label, usually short-lived, and generally, overlap with other tasks. One such problem pertains to bistable perception, which we consider to consist of two internal mental states, namely, transition and maintenance. The transition state is short-lived and represents a change in perception while the maintenance state is comparatively longer and represents a stable perception. In this study, we proposed a novel approach for characterizing the duration of transition and maintenance states and classified them from the neuromagnetic brain responses. Participants were presented with various types of ambiguous visual stimuli on which they indicated the moments of perceptual switches, while their magnetoencephalogram (MEG) data were recorded. We extracted different spatio-temporal features based on wavelet transform, and classified transition and maintenance states on a trial-by-trial basis. We obtained a classification accuracy of 79.58% and 78.40% using SVM and ANN classifiers, respectively. Next, we investigated the temporal fluctuations of these internal mental representations as captured by our classifier model and found that the accuracy showed a decreasing trend as the maintenance state was moved towards the next transition state. Further, to identify the neural sources corresponding to these internal mental states, we performed source analysis on MEG signals. We observed the involvement of sources from the parietal lobe, occipital lobe, and cerebellum in distinguishing transition and maintenance states. Cross-conditional classification analysis established generalization potential of wavelet features. Altogether, this study presents an automatic classification of endogenous mental states involved in bistable perception by establishing brain-behavior relationships at the single-trial level.

Original languageEnglish
Pages (from-to)95-113
Number of pages19
JournalCognitive Neurodynamics
Issue number1
Publication statusPublished - 2020 Feb 1


  • ANN
  • Bistable perception
  • Internal mental states
  • MEG
  • SVM
  • Single-trial classification
  • Source reconstruction

ASJC Scopus subject areas

  • Cognitive Neuroscience


Dive into the research topics of 'Switch or stay? Automatic classification of internal mental states in bistable perception'. Together they form a unique fingerprint.

Cite this