Introduction

Eye tracking captures gaze information in the form of fixations and saccades. Fixations occur when subjects focus their vision on a point in space (usually a screen) over time. Fixation count, rate, and duration are measured to reflect attention fixation and stimulation time [1, 2]. In contrast, saccades are quick shifts in eye position. Analyzing saccade angle and scan paths is helpful in distinguishing both stimulus-driven and automatic shifts in attention and executive function [3, 4]. Moreover, response latency, response time, and kinematics can be measured from eye movements [5,6,7]. An increasing number of studies are showing that eye tracking metrics not only measure basic oculomotor characteristics but also reflect complex cognitive information and predict specific cognition impairments. These metrics differ based on an individual’s perceptual characteristics, cognitive skills, and cognitive patterns. It has been identified that multiply brain areas including fronto-insular cortex, anterior cingulate cortex, supplementary motor area, superior colliculi, and thalamus can be activated when performing fixation and smooth pursuit tasks [8], while the bilateral dorsolateral prefrontal cortex has been involved in fixation durations [9]. Executive function involves the multiple cortical and subcortical regions, which are activated in saccade, smooth pursuit, visual searching, and social cognition tasks [10]. Therefore, eye tracking metrics provide abundant sources of data to study the bridge among behavior, brain function, and neural mechanisms, the inner workings of the mind [11, 12].

Over the past four decades, eye tracking technology has gained attention for its potential use in screening for cognitive dysfunction in neurological disorders. Thickbroom and Black first detected oculomotor abnormalities in multiple sclerosis (MS) using an electrooculographic system to measure eye motion in a target pursuit task [13]. In a separate study, 11 of 18 patients (61%) with amyotrophic lateral sclerosis (ALS) were found to have impaired pursuit eye movement, suggesting that this pursuit defect is a sign of extrapyramidal or supratentorial pyramidal involvement [14]. These findings opened the door for eye tracking to be developed as a diagnostic tool and as a marker for disease progression and prognosis.

It is known that many neurological disorders are accompanied by cognitive impairment, which requires early cognitive evaluation and long-term clinical monitoring [15]. Although conventional metrics such as cognitive assessment scales are widely used in clinical settings [16], their limitations are significant: patient evaluation requires intensive labor, result grading has poor resolution (e.g., absent/mild/moderate), and there is no iterative feedback based on large datasets [17]. Moreover, for testers’ results to be considered reliable, they would require intensive training [18]. In contrast, eye tracking can provide millisecond-level resolution and quantified parameters such as amplitude, latency, frequency, and stability [1, 19], which capture more objective, reliable, and scalable dynamic nature of behavior in natural environments [20]. Large datasets collected through eye tracking allow further analysis with sophisticated methods in machine learning [20]. Increasing evidence show that eye tracking information correlates well with traditional cognitive assessment scales, strongly suggesting that eye tracking can be used to evaluate and monitor cognitive states, disease severity, and disease progression in neurological diseases [1].

Methods

Search strategy

PubMed/MEDLINE and Web of Science were used via EndNote X8 in August 2018. The following terms were used:[((eye task) OR (eye AND task) OR (eye movement) OR (eye AND movement) OR (eye tracking) OR (eye AND tracking) OR (oculomotor) OR (cognition) OR (neurology) OR (neurological) OR (Amyotrophic lateral sclerosis) OR (ALS) OR (Alzheimer’s disease) OR (AD) OR (Parkinson’s disease) OR (PD) OR (Multiple sclerosis) OR (MS) OR (Epilepsy))]. There was no date limit in database searches. All medical subject headings (MeSH) terms were exploded to expand our search for related research. Some references related to relevant reviews and empirical articles were accessed for further investigation.

Study selection criteria

Included studies include

  • Patients with a diagnosis of neurological diseases.

  • A healthy control group.

  • The use of eye tasks with eye trackers to measure cognitive impairment: saccadic tasks, smooth pursuit tasks, fixation tasks, vergence tasks.

  • Reported original research.

  • Publication in English.

  • Publication in a peer-reviewed journal.

Excluded studies include

  • Single case studies.

  • Studies that came to a non-clinical outcome (i.e., reviews or studies validating eye movement measures).

  • Studies that used interviews, behavioral tasks, or questionnaires to examine eye movement.

  • Studies that did not report adequate data to illustrate a credible effect.

Methods of review

Figure 1 displays the process of paper selection for this review. The preliminary search retrieved 1287 articles, which became 1080 articles after duplication removal. One reviewer screened all titles and abstracts while another independent reviewer screened all articles to ensure inter-rater reliability, which was 84%. We accessed the article if it clearly met the inclusion criteria outlined above. To prevent omission, we retained the article when the reviewers could not determine with certainty whether the article met the inclusion criteria or not. Full manuscripts of 1080 studies were obtained in this way. Six additional papers were identified by manually searching the reference lists of these papers as well as two additional relevant reviews identified in the search. After excluding 992 records, the manuscripts of the 88 studies meeting inclusion criteria were reviewed by two independent raters. The differences were resolved through discussion and consultation. Forty-three articles were excluded after the final screening, 30 of which did not include eye tracking and 13 used eye tracking but met other exclusion criteria: case studies (n = 4), no control group (n = 2), data not sufficient to detect an effect (n = 3), and non-clinical outcome studies (n = 4).

Fig. 1
figure 1

Systematic review search process

Full size image

Discussion

Eye tracking tasks can be divided into five basic types: saccades, fixation, smooth pursuit, visual searching, and social cognition. Among them, saccades and fixation are most widely used tasks. Saccades comprise of prosaccades, antisaccades, and remembered saccades [21]. Prosaccades are fast eye movements towards a target, which can be initiated reflexively (externally cued)—such as when a bright light in the corner of the room elicits an automatic saccade towards that corner—or intentionally (internally driven, “voluntary”)—such as asking a subject to glance at the corner of the screen where a face appears. Antisaccades are saccades in the direction opposite to a peripheral visual target [21]. Remembered saccades are saccades towards a peripheral visual target that is no longer present [22]. When performing a saccadic task, participants are required to make a saccade as soon as the target appears on screen, during which he or she is recorded on an eye tracking system [23]. The voluntary saccade response time, the latency to switch from a prosaccade to an antisaccade, and the antisaccadic error rate are usually analyzed [24]. These parameters of saccades mainly reflect the impairment of executive function, further reflecting the related cortical and subcortical dysfunctions. Fixation tasks usually require subjects to look at a target without blinking for around 10 s while the fixation time and frequency would be measured to predict visual and attention impairment, implying damage of anterior cingulate and frontal cortices [2]. In smooth pursuit tasks, participants are required to follow a moving target with their eyes and the error rate will be analyzed [2]. Visual search focuses more on identifying a primed item from distractors, and the accuracy will be analyzed to evaluate executive dysfunction [25], while the facial emotions are required to recognize when performing a social cognition task [26]. Considering characteristics of ocular movements in different neurological disorders [8], researchers have to design corresponding eye tracking tasks for evaluating and monitoring disease progression in cognitively impaired patients.

Amyotrophic lateral sclerosis

ALS is a progressive paralytic disorder characterized by degeneration of motor neurons controlling voluntary muscles [27]. In addition to motor impairment, ALS patients often display behavioral and cognitive deficits, especially in the executive domain [28, 29]. They display higher antisaccade error rate due to the failure to suppress reflexive saccadic eye movements [6, 30]. Increased proportion of early saccades and slowed reflexive saccades have also been reported [31]. Moreover, ALS patients were impaired at emotion recognition with longer thinking time related to poorer performance [26]. Conventional “paper and pencil” tests are widely used in the evaluation of ALS (e.g., the Edinburgh Cognitive and Behavioral ALS Screen (ECAS)) [6, 30]. However, in moderate to severe stages of ALS, patients lose the ability to speak or write as a result of lower motor neuron atrophy, at which time traditional scales are no longer suitable for cognitive assessment [32].

Eye tracking measures were tested alongside traditional assessment scales with the hope that they can supplement or perhaps even supplant the latter during advanced disease states. Keller found that when ALS patients were asked to do ECAS tasks by directing their gaze at the answer displayed on a screen, ECAS testing time was significantly shortened [6], demonstrating that eye tracking could improve ECAS evaluation efficiency. The results also showed that ALS patients performed worse in the eye tracking version of the ECAS in the executive domain, as expected from our knowledge of ALS cognitive impairment. Proudfoot and Witiuk designed prosaccade and antisaccade tasks in which participants were asked to saccade towards (prosaccade) or away from (antisaccade) a target while their eyes were being tracked. Both studies also showed executive dysfunction in ALS due to higher antisaccade error rate and increased saccadic latency [25, 31]. In addition, poor performance in social cognition was also found when combined with emotion recognition tasks in ALS patients. Poletti adapted the Reading the Mind in the Eyes test, which challenged participants to derive emotion from pictures of eyes, to include an eye tracking component. They found that ALS patients exhibited longer mean latency and had fewer correct responses than healthy controls [30], highlighting social cognitive deficits in ALS patients. A similar experiment conducted by Girardi used gaze tracking during a social and emotional cognition task in which participants were asked to select the correct emotion shown on a picture of a face [26]. The ALS group produced fewer correct responses than healthy controls, supporting the finding of impaired emotion recognition and executive function in ALS.

These studies provide evidence that eye tracking can be used as a fast and reliable method for cognitive assessment [6] in patients with upper limb weakness or bulbar dysfunction [25]. However, in patients with highly advanced ALS with complete loss of voluntary eye movement [25], tools based on brain-computer-interface control systems instead of oculomotor measurements would be needed [6, 30, 33].

Alzheimer’s disease

Alzheimer’s disease (AD) is the most common neurodegenerative dementia, characterized by progressive memory loss, impaired attention, and executive dysfunction [2]. Oculomotor deficits in AD include saccade, fixation, and smooth pursuing. Increased large intrusive saccades and less accurate saccadic movements of AD patients were found in saccadic tasks [34]. In fixation tasks, AD patients required a longer amount of time to fixate the target but had shorter fixation duration [2]. When performing smooth pursuit tasks, they spent less time pursuing the target and had higher error rate [2]. Early diagnoses of cognitive deficits are critical for early intervention and improved prognosis [35]. Despite much progress in traditional cognitive assessment scales [36,37,38,39,40], sensitive markers are needed to facilitate earlier diagnosis and to serve as an outcome measure in clinical trials [41]. Early in 1992, Daffner applied oculomotor tracking to measure the response to provocative visual stimuli in AD patients and found that AD patients exhibited diminished visual curiosity compared with controls [42]. Since then, eye tracking has been proposed as a non-verbal and non-invasive paradigm for cognitive assessment and disease progression prediction in AD [37, 43].

To date, fixation, saccadic, and smooth pursuit tasks are the most common eye tracking tasks for AD [2]. During a saccadic task in AD, oculomotor movement is commonly recorded by a binocular infrared eye tracking system [44,45,46,47]. Parameters such as fixation duration, reaction time, saccadic latency, and saccadic error rate are analyzed. Compared with healthy controls, AD patients demonstrated shorter fixation periods [2], less accurate prosaccades, longer latency time for saccade initiation, and a higher number of saccades [2, 34, 47], conveying selective and executive attention impairment [34]. Moreover, AD patients spent less time following a target during a smooth pursuit task [2], suggesting that they have complex visual processing deficits due to cortical-subcortical disturbance [2, 48]. In a social and emotional cognition study, apathetic AD patients displayed reduced fixation duration and fixation frequency on social images, showing impairment in these areas [49]. Last, AD patients showed less preference for novel images than healthy controls, as evidenced by reduced looking time and fixation frequency [46]. In fact, declined visual-selective attention towards novel stimuli may be a characteristic cognitive deficit in AD patients.

Saccadic tasks may not be suitable for every AD patient or the same cognitive process, as performance may be influenced by a patient’s personal visual preferences [45]. Future work should focus on exploring advanced eye tracking tasks and parameters to evaluate more specific cognitive functions such as cognitive flexibility, short-term memory, and selective attention [34].

Parkinson’s disease

PD is also one of the most important neurodegenerative disorders of the central nervous system affecting the motor function. PD patients report cognitive decline particularly executive dysfunction as one of their greatest complaints. They often display oculomotor abnormalities during saccade tasks with higher prosaccadic latency and antisaccadic error rates, accompanied by increased disinhibitions [3]. Mild cognitive impairment and dementia affect up to 80% of PD [50]. Neuropsychological assessment scales like the Outcomes of Parkinson’s Disease—Cognition (SCOPA-COG) and the Parkinson’s Disease—Cognitive Rating Scale (PD-CRS) [51] have been widely used in clinical settings to detect cognitive impairment in PD [24]. Considering the low sensitivity of traditional scales to recognize subtle cognitive deficits in the early stages of PD [52], attention is shifting towards advanced neurophysiological measures such as pupillometry [24].

Similar to AD, saccadic tasks are the most widely used eye tracking tasks in PD cognitive evaluation due to its characteristic extraocular dystonia [3, 10, 21,22,23,24, 53, 54]. Executive dysfunction in PD is associated with a higher antisaccade error rate, an increased number of disinhibitions in delayed antisaccade tasks, and a longer response time of saccades [21]. Eye tracking metrics correlate with disease severity, suggesting that eye tracking measurement may predict disease progression in PD patients with cognitive impairment [47]. Combinations of the oculomotor version of the Simon task and the Barratt Impulsiveness Scale (BIS) questionnaire were used to investigate executive dysfunction in PD patients [53]. The results showed that, whereas fluency task scores failed to distinguish cognitive differences between early-stage PD patients and healthy controls, prosaccadic latency correlated strongly with basic oculomotor characteristics and predicted executive function impairment. In a separate study, a battery of tests were performed in drug-naive PD patients and healthy controls [10]. Newly diagnosed and unmedicated PD patients exhibited higher antisaccadic error rates and switch costs in the task switching test and performed significantly worse in the rule finding task, suggesting that abnormalities in saccadic behavior could be an early sign of cognitive impairment in PD. Similar results were obtained by Amador and Clark [21, 55].

However, no difference was found in the dynamics of saccadic eye movements in PD patients versus controls in some studies, which was explained as a result of the skeletomotor loop passing through the basal ganglia, independent of the oculomotor loop [3]. Other researchers argue that studies measuring eye movement characteristics alone are insufficient to define a full paradigm for cognitive function in PD. Multi-domain tasks such as the dual-target reach task which involves hand-eye coordination could quantify disease-specific cognitive deficits and may serve as a better tool to monitor PD progression. In addition, in most of the mentioned studies, the majority of PD patients were on dopaminergic medication during testing, which may have some effect on eye movement. Future studies should remove this confounding effect by testing both medicated and unmedicated PD patients [47].

Multiple sclerosis

MS is a demyelinating disease of the central nervous system characterized by diffuse tissue damage to white and gray matter regions [56]. Impaired mobility and cognition typically emerge 10 to 20 years after initial presentation. Impairment of attention, executive function, and memory are increasingly being recognized as important functional disabilities in MS [57]. Oculomotor dysfunction in MS, including fixation instability, higher saccade error rates, and impaired pursuit, has been identified commonly due to lesions of either the brainstem or the eye fields [58]. Traditionally, magnetic resonance imaging (MRI) of the brain [59] and neurophysiological tests [60] are used as diagnostic and prognostic markers in MS-related cognitive decline. However, regular MRI cannot be used directly in cognitive evaluation and assessment scales are time-consuming and subjective. For example, with regard to the most widely used and gold standard scale for cognitive assessment in MS, the Minimal Assessment of Cognitive Function in Multiple Sclerosis (MACFIMS), even a trained evaluator needs at least 90 min for a full evaluation, as a result MS patients often cannot cooperate [61]. Clearly, a faster and more accurate method to evaluate the complex form of cognitive dysfunction in MS is needed.

Eye tracking has been used to screen for abnormal visuospatial behavior in MS [5]. Due to its close association with ocular nerve impairment, saccadic tasks are most commonly used for oculomotor function assessment in MS patients [56, 62]. Clough designed an ocular working memory task in which participants were instructed to recall the positions of numerical stimuli on a screen, during the process of recording by an eye tracker [63]. The results showed higher error rates in the MS group, with more working memory errors made by patients who were farther along in the disease course, in line with previous studies [63,64,65]. Fielding conducted a battery of unpredictable, predictable, and endogenously cued visually guided saccade tasks on MS patients [62]. Participants were asked to look directly at the center of a green target cross, then pursue the target as it moved horizontally across the screen while ignoring a visual distractor. Saccadic latency and absolute position error were measured. Compared with a healthy group, MS patients exhibited increased saccadic latency [5, 56, 62, 66], worse fixation, and more prosaccade errors [66]. These results demonstrate the diagnostic and prognostic potential of eye tracking to assess cognitive function and disease severity in MS [56].

Epilepsy

Epilepsy is a chronic neurological disorder characterized by epileptic seizures [67], and usually accompanied by neuropsychological impairment at disease onset or even prior to it [11]. Patients with epilepsy were impaired at saccade in oculomotor function [11], with increased error rate when performing vision-guided saccade, pro saccade and antisaccade, and with longer reflexive time at the initial of saccade [68]. Unlike other disorders discussed above, there is no standard cognitive assessment tool for epilepsy. Seizure-related cognitive evaluation has mostly been performed on animal models such as rats and mice [69]. Even though cognitive scales have been designed for individuals with epilepsy, such as Epitrack and Portland Neurotoxicity Scale, they have several shortcomings including limited sensitivity, unsuitability for repeated assessment, and sole focus on one aspect of cognition, thus limiting their application [70].

Previous studies suggested that cognitive dysfunction in epilepsy has multifactorial determinants including the presence of structural lesions, seizures, interictal epileptic activity, drug treatments, psychiatric comorbidities, and individual reserve capacity. Therefore, it is almost impossible to retrospectively attribute cognitive deficits to a few specific factors, which makes early neuropsychological diagnostics critical [67, 68, 71,72,73]. A multifactorial and neurodevelopmental model for cognitive assessment in epilepsy patients is needed. Because oculomotor testing can assess response inhibition and working memory through related tasks [11], some researchers have suggested eye movement as a marker of cognitive impairment in epileptic patients [72].

Oculomotor tracking techniques applied to epilepsy can be used for two main purposes: clinical evaluation and the elucidation of neural mechanisms. During clinical evaluation, infrared eye trackers are used to record oculomotor parameters when performing vision-guided saccade, antisaccade response inhibition [11], prosaccade, and antisaccade tasks [68]. Okruszek found a correlation between emotional preference and memory activity in epileptic patients using the emotional faces recognition task [72]. Nagasawa and Lunn performed a voluntary vision-guided eye movement task and a word-reading task in children and adults with focal epilepsy [68, 74], respectively, which found that patients with a long course of epilepsy had higher antisaccade peak velocity and gain.

It is a fairly new approach to use oculomotor techniques for exploring neural mechanisms in epilepsy. Nagasawa used eye tracking to investigate the mechanism for cognitive deficits in epilepsy, which was followed by several other research groups [12, 71, 73, 74]. Oculomotor response and intracranial electroencephalogram (EEG) were recorded while patients conducted saccadic tasks, pursuit tasks [12], or emotional face recognition tasks [72]. Direct EEG recordings of the pursuit system showed latency of increased gamma right before target onset in the frontal eye field and the ventral intraparietal sulcus, revealing functional dissociation between these two regions [12]. Through a voluntary visually guided eye movement task, Nagasawa found that gamma-augmentation was seen in different areas of the brain such as the superior parietal lobule and the Rolandic area during perception of target motion and subsequent eye movement [74]. These results suggest that in-task eye movement patterns may serve as a marker of brain activation.

Conclusion

Diverse applications of eye tracking for cognitive evaluation in neurological disorders have drawn attention to the complexity of this technology, with a variety of tasks and analysis methods. Although there have been substantial advances in assessing cognitive impairment using eye tracking, the high financial cost of implementing these systems and the lack of evidence-based research must be addressed. This review discussed the available evidence on oculomotor characteristics of patients with neurological disorders and explored possibilities in future development into clinical applications. Eye tracking approaches can help expand our understanding of cognition and behavior in neurological disorders, leading to improvements in early diagnosis, long-term care, and the treatment of this important comorbidity.