Bayesian microsaccade detection

Mihali, Andra; Opheusden, Bas van; Ji, Wei

doi:10.1167/17.1.13

Cited by 107 publications

(28 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently, a Bayesian approach for the detection of microsaccades based on a generative model has been proposed 43 . Inherently, Bayesian methods provide estimates of uncertainty, in addition to estimates of the quantity of interest.…”

Section: Discussionmentioning

confidence: 99%

Human-level saccade detection performance using deep neural networks

Bellet

Nienborg

et al. 2018

Preprint

View full text Add to dashboard Cite

Saccades are ballistic eye movements that rapidly shift gaze from one location of visual space to another. Detecting saccades in eye movement recordings is important not only for studying the neural mechanisms underlying sensory, motor, and cognitive processes, but also as a clinical and diagnostic tool. However, automatically detecting saccades can be difficult, particularly when such saccades are generated in coordination with other tracking eye movements, like smooth pursuits, or when the saccade amplitude is close to eye tracker noise levels, like with microsaccades. In such cases, labeling by human experts is required, but this is a tedious task prone to variability and error. We developed a convolutional neural network (CNN) to automatically detect saccades at human-level performance accuracy. Our algorithm surpasses state of the art according to common performance metrics, and will facilitate studies of neurophysiological processes underlying saccade generation and visual processing.

show abstract

Section: Discussionmentioning

confidence: 99%

Human-level saccade detection performance using deep neural networks

Bellet

Nienborg

et al. 2018

Preprint

View full text Add to dashboard Cite

show abstract

“…Artifact rejection, and some forms of artifact correction methods (Hassler et al, ; Keren et al, ), rely on accurate detection of the artifacts. The most reliable detection of saccades is achieved using a high‐resolution eye tracker combined with a detection algorithm (Engbert & Kliegl, ; Engbert & Mergenthaler, ; Mihali, van Opheusden, & Ma, ; Otero‐Millan, Castro, Macknik, & Martinez‐Conde, ). However, high‐resolution eye trackers are relatively expensive and not always available.…”

Section: Discussionmentioning

confidence: 99%

Reducing saccadic artifacts and confounds in brain imaging studies through experimental design

Tal

Yuval-Greenberg

2018

Psychophysiology

View full text Add to dashboard Cite

Saccades constitute a major source of artifacts and confounds in brain imaging studies. Whereas some artifacts can be removed by omitting segments of data, saccadic artifacts cannot be typically eliminated by this method because of their high occurrence rate even during fixation (1-3 per second). Some saccadic artifacts can be alleviated by offline-correction algorithms, but these methods leave nonnegligible residuals and cannot mitigate the saccade-related visual activity. Here, we propose a novel yet simple approach for diminishing saccadic artifacts and confounds through experimental design. We suggest that specific tasks can lead to substantially less saccade occurrences around the time of stimulus presentation, starting from slightly before its onset and lasting for a few hundred milliseconds. In three experiments, we compared the frequency and size of saccades in a variety of tasks. Results of Experiment 1 showed that a foveal change-detection task reduced the number and sizes of saccades, relative to a parafoveal orientation-discrimination task. Experiment 2 replicated this finding with a parafoveal object recognition task. Experiment 3 showed that both foveal and parafoveal continuous change detection tasks induced fewer and smaller saccades than a discrete orientation-discrimination task. We conclude that adding a foveal or a parafoveal continuous task reduces saccades' number and size. This would lead to better artifact correction and enable the omission of contaminated data segments. This study may be the first step toward developing saccade-free experimental designs.

show abstract

“…That is, we can parameterize data differences in terms of r w , r v , and h before producing the detection outcome. Comparison with the BIM We now perform the comparison with BIM using dataset in Mihali et al (2017). It contains five sequences of tracked gaze positions from five participants, and Table 4 shows the detail.…”

Section: Comparison With the Cnn Methodsmentioning

confidence: 99%

“…Particle filtering, with a small number of effective particles, was proposed to detect (micro)saccades (Daye & Optican, 2014). More recently, the hidden Markov model was proposed in Mihali, van Opheusden, and Ma (2017), and the Bayesian Inference Method (BIM) was derived accordingly as the microsaccade detector. A reliable and robust detection method would inevitably rely on an accurate statistical model.…”

Section: Introductionmentioning

confidence: 99%

“…Therefore, the calculation of the minimum variance guarantees that the estimation of the variance is from a subsection of the state H 0 , and not H 1 . Figure 3 shows the estimated nr 2 w þ 2r 2 v with respect to different n values using Algorithm 1 with the real gaze data provided in Mihali et al (2017). The dotted line in Figure 3a depicts the estimated variance from the dataset, and the dashed line in Figure 3a shows the theoretical variance from fitted data points.…”

mentioning

confidence: 99%

See 1 more Smart Citation

The Neyman Pearson detection of microsaccades with maximum likelihood estimation of parameters

2019

View full text Add to dashboard Cite

# Despite the fact that the velocity threshold method is widely applied, the detection of microsaccades continues to be a challenging problem, due to gaze-tracking inaccuracy and the transient nature of microsaccades. Important parameters associated with a saccadic event, e.g., saccade duration, amplitude, and maximum velocity, are sometimes imprecisely estimated, which may lead to biases in inferring the roles of microsaccades in perception and cognition. To overcome the biases and have a better detection algorithm for microsaccades, we propose a novel statistical model for the tracked gaze positions during eye fixations. In this model, we incorporate a parametrization that has been previously applied to model saccades, which allows us to veridically capture the velocity profile of saccadic eye movements. Based on our model, we derive the Neyman Pearson Detector (NPD) for saccadic events. Implemented in conjunction with the maximum likelihood estimation method, our NPD can detect a saccadic event and estimate all parameters simultaneously. Because of its adaptive nature and its statistical optimality, our NPD method was able to better detect microsaccades in some datasets when compared with a recently proposed stateof-the-art method based on convolutional neural networks. NPD also yielded comparable performance with a recently developed Bayesian algorithm, with the added benefit of modeling a more biologically veridical velocity profile of the saccade. As opposed to these algorithms, NPD can lend itself better to online saccade detection, and thus has potential for human-computer interaction applications. Our algorithm is publicly available at https://github.com/hz-zhu/NPD-microsaccade-detection.

show abstract

Bayesian microsaccade detection

Cited by 107 publications

References 48 publications

Human-level saccade detection performance using deep neural networks

Human-level saccade detection performance using deep neural networks

Reducing saccadic artifacts and confounds in brain imaging studies through experimental design

The Neyman Pearson detection of microsaccades with maximum likelihood estimation of parameters

Contact Info

Product

Resources

About