The present study aims to apply multiscale entropy (MSE) to analyse brain activity in terms of brain complexity levels and to use simultaneous electroencephalogram and functional near-infrared spectroscopy (EEG/fNIRS) recordings for brain functional analysis. A memory task was selected to demonstrate the potential of this multimodality approach since memory is a highly complex neurocognitive process, and the mechanisms governing selective retention of memories are not fully understood by other approaches. In this study, 15 healthy participants with normal colour vision participated in the visual memory task, which involved the making the executive decision of remembering or forgetting the visual stimuli based on his/her own will. In a continuous stimulus set, 250 indoor/outdoor scenes were presented at random, between periods of fixation on a black background. The participants were instructed to make a binary choice indicating whether they wished to remember or forget the image; both stimulus and response times were stored for analysis. The participants then performed a scene recognition test to confirm whether or not they remembered the images. The results revealed that the participants intentionally memorising a visual scene demonstrate significantly greater brain complexity levels in the prefrontal and frontal lobe than when purposefully forgetting a scene; p < 0.05 (two-tailed). This suggests that simultaneous EEG and fNIRS can be used for brain functional analysis, and MSE might be the potential indicator for this multimodality approach.
Mismatch negativity (MMN) is a component of the difference waveform derived from passive auditory oddball stimulation. Since its inception in 1978, this has become one of the most popular event-related potential techniques, with over two-thousand published studies using this method. This is a testament to the ingenuity and commitment of generations of researchers engaging in basic, clinical and animal research. Despite this intensive effort, high-level descriptions of the mechanisms theorized to underpin mismatch negativity have scarcely changed over the past four decades. The prevailing deviance detection theory posits that MMN reflects inattentive detection of difference between repetitive standard and infrequent deviant stimuli due to a mismatch between the unexpected deviant and a memory representation of the standard. Evidence for these mechanisms is inconclusive, and a plausible alternative sensory processing theory considers fundamental principles of sensory neurophysiology to be the primary source of differences between standard and deviant responses evoked during passive oddball stimulation. By frequently being restated without appropriate methods to exclude alternatives, the potentially flawed deviance detection theory has remained largely dominant, which could lead some researchers and clinicians to assume its veracity implicitly. It is important to have a more comprehensive understanding of the source(s) of MMN generation before its widespread application as a clinical biomarker. This review evaluates issues of validity concerning the prevailing theoretical account of mismatch negativity and the passive auditory oddball paradigm, highlighting several limitations regarding its interpretation and clinical application.
1 December 2020 in European Journal of Neuroscience
Gliomas observed in medical images require expert neuro-radiologist evaluation for treatment planning and monitoring, motivating development of intelligent systems capable of automating aspects of tumour evaluation. Deep learning models for automatic image segmentation rely on the amount and quality of training data. In this study we developed a neuroimaging synthesis technique to augment data for training fully-convolutional networks (U-nets) to perform automatic glioma segmentation. We used StyleGAN2-ada to simultaneously generate fluid-attenuated inversion recovery (FLAIR) magnetic resonance images and corresponding glioma segmentation masks. Synthetic data were successively added to real training data (n = 2751) in fourteen rounds of 1000 and used to train U-nets that were evaluated on held-out validation (n = 590) and test sets (n = 588). U-nets were trained with and without geometric augmentation (translation, zoom and shear), and Dice coefficients were computed to evaluate segmentation performance. We also monitored the number of training iterations before stopping, total training time, and time per iteration to evaluate computational costs associated with training each U-net. Synthetic data augmentation yielded marginal improvements in Dice coefficients (validation set +0.0409, test set +0.0355), whereas geometric augmentation improved generalization (standard deviation between training, validation and test set performances of 0.01 with, and 0.04 without geometric augmentation). Based on the modest performance gains for automatic glioma segmentation we find it hard to justify the computational expense of developing a synthetic image generation pipeline. Future work may seek to optimize the efficiency of synthetic data generation for augmentation of neuroimaging data.