Submitted Successfully!
To reward your contribution, here is a gift for you: A free trial for our video production service.
Thank you for your contribution! You can also upload a video entry or images related to this topic.
Version Summary Created by Modification Content Size Created at Operation
1 -- 4820 2023-02-21 21:15:16 |
2 format + 1 word(s) 4821 2023-02-22 03:18:25 |

Video Upload Options

Do you have a full video?


Are you sure to Delete?
If you have any further questions, please contact Encyclopedia Editorial Office.
Przybyszewski, A.W.; Śledzianowski, A.; Chudzik, A.; Szlufik, S.; Koziorowski, D. Machine Learning and Eye Movements. Encyclopedia. Available online: (accessed on 23 June 2024).
Przybyszewski AW, Śledzianowski A, Chudzik A, Szlufik S, Koziorowski D. Machine Learning and Eye Movements. Encyclopedia. Available at: Accessed June 23, 2024.
Przybyszewski, Andrzej W., Albert Śledzianowski, Artur Chudzik, Stanisław Szlufik, Dariusz Koziorowski. "Machine Learning and Eye Movements" Encyclopedia, (accessed June 23, 2024).
Przybyszewski, A.W., Śledzianowski, A., Chudzik, A., Szlufik, S., & Koziorowski, D. (2023, February 21). Machine Learning and Eye Movements. In Encyclopedia.
Przybyszewski, Andrzej W., et al. "Machine Learning and Eye Movements." Encyclopedia. Web. 21 February, 2023.
Machine Learning and Eye Movements

Humans are a vision-dominated species; what we perceive depends on where we look. Therefore, eye movements (EMs) are essential to our interactions with the environment, and experimental findings show EMs are affected in neurodegenerative disorders (ND). This could be a reason for some cognitive and movement disorders in ND. Therefore, several research aim to establish whether changes in EM-evoked responses can tell us about the progression of ND, such as Alzheimer’s (AD) and Parkinson’s diseases (PD), in different stages.

Alzheimer’s disease Parkinson’s disease eye movements

1. Standard Neurological Approach

Experienced neurologists use their clinical knowledge and experience from many years of practice to estimate symptom development and the best treatment for an individual Parkinson’s disease (PD) patient. However, because of the long period of compensatory mechanisms unique to each patient, there is a famous saying that “No two people face Parkinson’s in quite the same way.” Therefore, a neurologist must consider not only motor symptoms, but many others, and must even try to understand a patient’s cognitive and emotional symptoms through the Theory of Mind space [1][2] to estimate disease progression in the individual patient.
Typically, eye movement analysis coexists in the broader context of other neuropsychological measurements. They can be related to the following tests: PDQ-39/PDQ-8—a 39/8-question test related to health difficulties in everyday living (summaries of both tests are strongly correlated), ESS—Epworth sleeping scale results (related to sleepiness problems, predominantly during the day), BDI—Beck’s depression inventory (21-item quantitative measure of symptoms of depression), TMT A and TMT B tests (Trail Making Test—part A measures psychomotor speed and part B is related to executive function). Another test, AIMS—Abnormal Involuntary Movement Score—measures the involuntary movements of patients. Additionally, neurologists can use more wide-ranging cognitive tests, such as MoCA (Montreal cognitive assessment) for the detection of MCI (mild cognitive impairment) for PD and AD or dementia (AD), or the similar, but shorter, MMSE (Mini-Mental State Examination), which is less sensitive than MoCA, but has established clinical values. Another important test, the FAS (Phonemic Verbal Fluency) test involves orally producing words that start with the letters F, A, and S, evaluates the cognitive function and measures language-related executive functioning. The most common decisive attribute is the UPDRS (Unified Parkinson’s Disease Rating Scale), an essential neurological test for the effects of PD long-treatment effects. The UPDRS score estimates daily activity (non-motor—UPDRS I and motor—UPDRS II) and motor-related problems (UPDRS III). It is the so-called “Gold Standard” for determining PD progressions. Alzheimer’s disease progression is mainly related to cognitive changes (normal, MCI, dementia), and disease progression is determined by the CDR (Cognitive Dementia Rating) scale as the decision attribute.

2. EM in PD—Saccades

There is pervasive, clinically oriented, literature related to reflexive saccade (RS) latency in PD in comparison to same-age healthy controls showing different, contradictory results [3]. However, a meta-analytic review [4] demonstrated, by analyzing 47 representative studies (from 1529 references), that the RS latency depends on which method was used by different authors: gap, step, or overlap. Different methods are determined by the time difference (∆t) between the fixation point disappearance and the target appearance: if ∆t > 0, then it is the gap method; if ∆t = 0, then it is the step method; and if ∆t < 0, it is the overlap method. By using quantitative pooling analysis, Chambers and Prescott [4] demonstrated that the slowed response time in PD, compared to a control, is strongest in the step method (∆t = 0), weaker in the gap method (∆t > 0), and negligible in the overlap method (∆t < 0).
In a study [5], standard neurological attributes and PDQ39, Epworth, and AIMS were measured, as well as EM—reflexive saccades. The authors compared different algorithms for the classification of UPDRS, UPDRS II, and UPDRS III in 10 PD patients. A six-fold cross-validation method was used. For the accuracy measure of UPDRS, the best results were given by RSES (Rough Set Exploration System based on Rough Set theory, which is an important implementation of GC—granular computing) with a global accuracy of 0.90, the second was Random Forest, which had a global accuracy of 0.68, and the third was a Tree Ensemble, with 0.65 global accuracy. For UPDRS II, the best results were for the Random Forest with an accuracy of 0.80, the second was RSES with 0.79, and the third was Bayesian classification, which gave an accuracy of 0.77. For the UPDRS III, the best was RSES classification, which gave a 0.82 accuracy, and the Decision Table (with Weka), which gave an accuracy of 0.77.
The results above agree in general with a recent publication [6] where they tested horizontal and vertical saccades and antisaccades (AS) for healthy control and PD subjects in different stages of the disease, measured by Hoehn and Yahr stages (H&Y2 and H&Y3). The PD group displayed decreased vertical saccade amplitudes and increased vertical saccade and AS latencies. AS latency increased for H&Y2 and H&Y3 patients, but AS errors (correlated with MoCA—Montreal cognitive assessment test’s score) were similar for control and H&Y2 subjects, but larger for H&Y3 subjects. Levodopa has increased vertical saccade latency but decreased AS latency. This work has described many different difficult-to-analyze mechanisms that might be easier to put through machine learning methods.
In one study [7], it was demonstrated that saccades could predict cognitive decline in PD patients. In 140 PD and 90 age-matched participants, the authors evaluated differences in RS metrics between early-PD and healthy age-matched adults. They assessed RS and cognition at baseline at 18, 36, and 54 months. RS parameters were latency, duration, amplitude, peak velocity, and average velocity, and the cognitive assessment contained executive function, attention, fluctuating attention, and memory. RS parameters, with the help of linear mixed-effects models, were used as predictors of cognitive decline over 54 months. At the baseline, RS was impaired in PD patients compared to the control group. RS parameters predicted a decline in global cognition, executive function (verbal fluency), attention, and memory over 54 months in PD patients. However, only reductions in global cognition and attention were predicted by RS parameters in age-matched subjects, which means that cognitive changes were not just age-related [7]. The dependence between RS latency and executive functions was also confirmed earlier [8]. In addition, manual and saccadic performances are uncorrelated in the average population, but both are similarly affected by PD [9].

3. EM in PD—Antisaccades

One of the well-proven experimental models used to examine the inhibition of automatic reflexive responses is the antisaccade task (AS) [10].
In a study [11], the significance of antisaccade (AS) parameters for the classification of Parkinson’s disease motor and motor variations (UPDRS II and UPDRS IV) was tested. There were 11 PD patients examined in 4 sessions. In addition to the standard neurological attributes, AS parameters such as delay, duration, and maximum speed were measured. RSES was used for the data discretization and attribute reduction and to perform a 5-fold cross-validation. The best result was obtained by the RSES Decomposition Tree, which splits the dataset into fragments represented as a tree’s leaves [11]. The UPDRS III classification results indicated an accuracy of 0.85 with a coverage of 0.48. Surprisingly, the UPDRS IV was estimated with an accuracy of 0.91 and coverage of 0.39, so UPDRS IV showed a more significant correlation with antisaccade parameters. Thus, UPDRS IV showed greater sensitivity in predicting antisaccade parameters [11]. From the results, it also emerged that attributes describing methods of patient treatment (again, the session attribute) and mean duration were most sensitive in predicting the scores of both UPDRS III and IV.
As described above, different analysis methods influence the RS latency [4]. In a review meta-analysis, Waldthaler et al. [12] analyzed the influence of the paradigm (gap, step, overlap) on AS latency and errors. They [12] compared the results of 703 PD patients with 600 healthy controls for antisaccade latency and 831 patients and 727 healthy controls for antisaccade error rate. Over 60% of studies excluded PD with dementia. Like RS latencies, the mean AS latency was 339.8 ms in the PD patients and 294.2 ms in the healthy group in the gap paradigm, and 411.7 ms in the PD patients and 368.6 ms in the healthy group in the step paradigm. This was measured for PD patients with disease duration between 0.7 and 14.7 years and UPDRS III scores between 5 and 85, from early to advanced disease stages. In a meta-analysis, the authors [12] demonstrated that AS latency increases with disease severity, and an increase in the levodopa dosage influences the AS error rate (negatively moderating effect).
A study by Waldthaler et al. [13] tested whether patients with Parkinson’s taking dopaminergic medication performed better at response inhibition during antisaccade tasks. Levodopa intake has favorable or harmful effects on dopamine-dependent cognitive tasks based on essential basal dopamine intensities in ventral segments of the striatum, agreeing with the dopamine overdose theory. Thirty-five patients with Parkinson’s (and 30 healthy subjects) completed antisaccade tasks in OFF and ON medication conditions. Investigators computed multiple linear regressions to forecast the alterations in antisaccade delay and directive mistakes, and to express saccade rate based on age at Parkinson’s disease onset, disease duration, levodopa-equivalent circadian amount, motor indicator difficulties, and executive functions. According to their results, earlier disease onset and milder motor symptoms in the OFF-medication status were related to diminished inhibition ability response after levodopa intake, mirrored in enlarged express saccades and mistakes. They concluded that levodopa might have opposite results on oculomotor reaction inhibition contingent on the age at Parkinson’s disease onset and motor disease gravity.
During their next study, Waldthaler et al. [14] examined whether there was any correlation between the development of motor and cognitive indications in 25 patients and Parkinson’s disease (age: 61.4 +/− 6.8, disease duration: 6.0 +/− 4.5 years). A total of 10 patients from all 25 PD patients received subthalamic nucleus DBS (deep brain stimulation) during the follow-up period (from DBS surgery to follow-up visit: 4.5 +/− 2.1 months). All PD patients were examined in ON medication and ON-DBS states, and modifications of dopaminergic treatment were permitted during the follow-up epoch. PD patients without DBS who displayed substantial improvement in motor signs after one year also received higher levodopa equivalent dosages at follow-up. Generally, the antisaccade (AS) delay (baseline: 339 +/− 72 ms, mean change: 95 +/− 1.1 ms) and mistake rate (baseline: 0.52, mean change: −0.02 +/− 0.3) stayed steady in the non-DBS group. In the DBS group, the AS delay tended to increase (baseline: 295 +/− 78 ms, mean change: 48 +/− 75 ms (p = 0.09)), but the mistake rate improved at follow-up (baseline: 0.76, mean change: −0.21 +/− 0.3 (p = 0.048)). The change in AS delay was connected to change in MDS-UPDRS III in both groups (non-DBS group baseline: 25.7 +/− 13, mean change: 0.3 +/− 7.4; DBS group baseline: 27.2 +/− 16.5, mean change: −5.2 +/− 17.8) and with the change in MoCA score in the non-DBS group (25.8 +/− 3.1, mean change 1.3 +/− 3.1). The authors indicate that AS delay may be sensitive to the development of motor and cognitive signs over time in Parkinson’s disease patients.

4. EM in PD—Saccades and Antisaccades

The same group of patients as in [11] was used for UPDRS prediction based on RS and AS measurements. The best accuracy of 0.89 was achieved by Decision Trees [15]. The results showed that the accuracy of the predictions increased with the number of significant attributes that were obtained by, for example, averaging RS and AS duration or by adding the averaged standard deviations of each patient’s latencies [15].
The authors of [12] demonstrated that RS and AS latencies were correlated with the results of neuropsychological tests in 65 PD patients, but only the results for AS latencies concerning patients’ cognitive impairment were statistically significant. In a study [16], 19 drug-naïve PD patients and 20 age-matched controls were examined. Patients had clinically probable idiopathic disease within three years of disease onset. Their RS latencies were like those of the controls, but AS error rates differed significantly (PD 15% vs. 8.7% for controls).
Fooken et al. [17] studied different tasks and conditions in which the oculomotor function in Parkinson’s patients is preserved. A total of 16 patients with Parkinson’s disease and 18 healthy, age-matched controls performed a set of tasks of saccades (RS), anti-saccades (AS), pursuits, and rapid ‘go/no-go’ manual interventions. Compared to the control group, PD patients showed regular impairment in tasks with fixed targets: prosaccades were hypometric, and AS were wrongly started towards the indicated target in 35% of the trials compared to 14% of errors in the control group. In PD subjects, task errors were linked with short-latency saccades, demonstrating anomalies in inhibitory control. However, the patients’ EMs in response to dynamic targets were well-preserved. Parkinson’s disease patients can track and predict a moving target and make quick go/no-go decisions with the same precision as healthy people. The intercepting hand movements of the patients were slower on average but indicated adaptive processes compensating for the motor slow down. Researchers concluded that the preservation of eye and hand movement functions in PD is linked to a separate functional pathway through the upper colliculus–brainstem loop that detours the frontal–basal ganglia network.
Kocoglu et al. [18] investigated how social processes and behaviors change in PD during spatial signaling tasks. Socially relevant directional cues, such as photos of people looking left or right, have been found to redirect attention. In conclusion, the basal ganglia can play a role in responding to such directional signals. In this research, patients and healthy controls performed pro- and anti-saccade tasks in which different directional signs preceded the appearance of the target. They analyzed reaction time, prediction errors, and correlations with PD severity and cognitive assessment scores. Patients displayed increased errors and answer times with the AS (antisaccade) task, but not with the RS (saccade) task. The control subjects made the most predictive errors in the finger-pointing trials, and the PD patients were mostly affected by the arrow, gaze, and pointing clues. It has been found that PD patients have a reduced ability to suppress responses to directional signals, but this effect is not specific to social signals.
Munoz et al. [19] studied whether bilateral deep stimulation of the basal ganglia–subthalamic nucleus (STN DBS) may affect the control of inhibition of eye movement in PD. They investigated the effect of DBS amplitude on inhibitory power during an antisaccade procedure on 10 PD patients after their DBS surgery. Subjects without medication (12 h, overnight) performed the antisaccade tasks with a set of different DBS stimulation amplitudes (from 0—no stimulation to 5—higher levels). The prosaccade error rate (related to a saccade at the beginning of the antisaccade) increased with increasing DBS stimulation amplitude (p < 0.01). Moreover, the saccade error rate increased with the decrease in the modeled volume of tissue activated (VTA) and decreased overlap of the STN stimulation area, but this connection was determined by the stimulation amplitude (p = 0.04). They concluded that the directional prosaccade error rate during the antisaccade task indicated impaired inhibitory control and suggested that higher stimulation amplitude settings can be modulatory for inhibitory control.

5. EM in PD—Pursuit

Another study tested how effective diagnostic parameters of slow (pursuit) eye movements are for the prediction of PD symptom development [20][21]. Horizontal pursuit EM with three different sinusoidal movement speeds was measured. The gain and accuracy (EM measurement section) were estimated. The discretization and attribute reduction with RSES demonstrated that the significant attributes were precise for the accuracy of the fastest sinusoidal movement speed, and gains decreased for medium and high sinusoidal movement light spot speeds [20][21]. The result of the 4-fold cross-validation gave a global accuracy of 0.77 for the UPDRS III prediction. An accuracy of 0.8 for the session number prediction (different treatments) in 10 PD patients was found. The above predictions were obtained for a sample of 20 patients using different binning methods (KNIME auto-binner), which allowed the grouping of UPDRS III data in intervals of equal frequencies. A 90% accuracy in predictions on these data was achieved with the RSES and 5-fold cross-validation [21]. When comparing the accuracy results of different classifiers, the RSES is in first place in the ranking, ahead of SVM (59%), Naive Bayes (55%), and Random Forest (52%) [21].
In this context, in her review, Frei [22] analyzed 29 articles (from 819 found) on smooth-pursuit eye movements in PD patients and compared them to those in normal subjects. She found that in 18 articles, the gain was measured and reduced in PD patients compared to controls in 16 of these papers. In two papers, the gain was reduced for higher target velocities. In three articles, accuracy was measured and found to be reduced in PD. There were also correcting saccades during smooth-pursuit EM that were more dominant in more advanced PD and for faster smooth pursuits, but quantification of saccades was difficult [22].
In another study, deep brain stimulation (DBS) increased smooth-pursuit accuracy (p < 0.001) and smooth-pursuit gain (p = 0.005), especially for faster smooth pursuits (p = 0.034) [23].
In their study, Farashi et al. [24] observed eye movements (EMs) during inactive states (eyes closed and eyes open), measuring EM using vertical electrooculography (VEOG). They performed the analysis in the time, frequency, and time–frequency axes of the VEOG time series. The authors completed a categorization by comparing healthy subjects and PD patients in OFF and ON medication conditions. They used an SVM (support vector machine) classifier and allowed multiple-differentiation-corrected p-values. The VEOG data achieved 69.10% and 87.27% discrimination precision for OFF and ON medication conditions, respectively. The authors established that PD patients’ vertical EM had smaller amplitude changes than healthy subjects in OFF medication conditions. The levodopa treatment augmented such changes in vertical EM during the eyes-closed situation and diminished during the eyes-open situation. As a result of levodopa treatment, VEOG time series amplitudes may change, although vertical EM rates were not affected (frequency contents).

6. EM in PD—Pupillometry

Parkinson’s disease patients develop a distorted pupillary response dependent on an abnormality in the retinal ganglion cells. Tabashum et al. [25] illustrated an arrangement for pupil size estimates that permits the discovery of pupil parameters to measure the post-illumination pupillary response (PIPR) with a Kalman filter estimating the pupil center and diameter over time. The pupillary reaction was estimated in the contralateral eye to two diverse light stimuli (470 and 610 nm) for 19 Parkinson’s patients and 10 healthy subjects. Net PIPR displayed different reactions to wavelengths (0.13 mm for Parkinson’s patients and 0.61 mm for healthy subjects, proving an extremely significant differentiation (p < 0.001)).
Tsitsi et al. [26] evaluated gaze constancy and pupil size in steady light surroundings, as well as eye movements (EMs) during constant fixation in a group of 50 Parkinson’s disease subjects (66% males) with unilateral to mild symptoms (Hoehn and Yahr 1–3; Schwab and England 70–90%) and 43 control subjects (37% males) with an eye tracker (1200 Hz) and logistic regression analysis. They examined the potency of the relationship of EM measures with the ROC curve results of 0.817, 95% CI: 0.732–0.901, and concluded that eye-tracking-established amounts of gaze fixation and pupil reaction might be valuable biomarkers of Parkinson’s disease indications.

7. EM in PD—Multimodal Approach

Bonnet et al. [27] investigated how connections between vision and posture are exaggerated in Parkinson’s patients. PD subjects have been shown to display unusually low levels of synergy in their posture self-control. These impaired reactions are related to the neurodegeneration processes in Parkinson’s disease that affect the basal ganglia, which facilitate the integration of both types of movements. They tested 20 PD patients (mean age: 60) on levodopa and 20 age-matched-healthy subjects (mean age: 61) with a detailed visual assignment (target-seeking scenarios in an image) and an inaccurate control task (arbitrarily viewing an image) in which pictures were projected onto a large screen. Lower back, upper back, head, and EM were registered simultaneously. To analyze behavioral synergies, the authors computed Pearson correlations between EM and postural actions. The associations between EM and upper- and lower-back movements were diminished in Parkinson’s subjects. The healthy control subjects did not display important correlations between EM and postural activities. Generally, their results revealed that the Parkinson’s subjects were unable to correct and change their postural rigidity to achieve success in the visual task. Moreover, these problems may occur in the early stages of Parkinson’s (an early biomarker opportunity).
Zhang et al. [28] investigated 49 Parkinson’s patients, including 35 early-stage (Hoehn and Yahr: 1–2 staging) and 14 advanced PD subjects (Hoehn and Yahr scale: 3 to 5 staging) and 23 healthy subjects. In addition to clinically significant PD symptoms, video-oculography was used to measure EM features such as eye fixation stability, horizontal and vertical reflexive saccade (RS), and horizontal and vertical smooth-pursuit movements. The authors discovered that five EM features—specifically square wave jerk frequency, vertical RS delays, the accuracy of the vertical–upward RS, and the horizontal smooth-pursuit RS gain—were meaningfully different in Parkinson’s and normal subjects. By merging all five features, the authors achieved a symptomatic sensitivity of 78.3% and a specificity of 95.2%. The study discovered that more deficiencies in upward–vertical RS than in other directions were related to disease duration and the stage of development of Parkinson’s disease.
Perkins et al. [29] investigated whether Sleep Behavior Disorder (RBD) indicates PD. With video-based eye tracking, researchers tested saccade, pupillary, and blink responses in RBD and isolated REM (rapid eye movement) with 22 PD and 22 RBD patients and 74 healthy controls. They found that RBD patients did not have significantly different saccades compared to healthy controls, but PD patients differed from both healthy controls and RBD patients. They concluded that RBD and PD patients had altered pupil and blink behavior compared to healthy controls. Because RBD saccade parameters were comparable to healthy controls, brain areas responsible for pupil and blink control may be impacted before saccadic control areas, making them a potential prodrome of PD.

8. Prediction of Disease Progression in Different PD Groups

The goal in [30] was to predict Parkinson’s disease progression in advanced-stage patients based on data obtained from patients under different treatments and at different stages of the disease. Patients from the BMT group (only on medication, third visit), DBS group (after recent deep brain stimulation surgery, third visit), and POP group (after older DBS surgery, first visit) were used as a training dataset—a model. The model was tested on the data obtained from the POP group during the second visit. A dedicated data science framework written in Python was used and based on the Scikit Learn and Pandas libraries that implemented different multiclass strategies, such as k-Nearest Neighbors Classifier, Support Vector Classifier, Decision Tree Classifier, and Random Forest Classifier. In this trial, the Random Forest Classifier achieved the highest overall accuracy score of 0.75 and an accuracy of 0.7 when predicting subclasses of UPDRS for patients in advanced stages of the disease who responded to treatment, with a global 0.57 accuracy score for all classes [30].
The purpose of another study [31] was to predict the results of different PD patient treatments to find the optimal one. The study compared the intelligent methods based on Rough Set theory with several different machine learning algorithms, namely Gaussian Naive Bayes, Decision Tree, Logistic Regression, C-Support Vector, Linear SVC, and Random Forest. Generally, the Rough Set method gave better accuracy, but less coverage, than other algorithms. On the other hand, the Rough Set-based approach allows the creation of more general rules without the necessity of additional data splitting (into different sessions), which was required in the other ML models to obtain accuracies similar to those obtained by RS. An example is the prediction of UPDRS in a DBS patient group from rules obtained from BMT patients. Global accuracy for DBS patients was 0.64 for the first visit, 0.85 for the second visit, and 0.74 for the third visit. Other methods gave accuracies of 0.88, 0.58, and 0.54, respectively [31].
The principal conclusion from this comparison is the observation that RS is a much more universal method when considering medical data. Finally, it was demonstrated that it is possible to estimate symptoms and their time development in populations treated differently, which may, in the future, lead to the discovery of universal rules of PD progression and to the optimization of treatment.

9. Prediction of Disease Progression Related to Motor, Cognitive, and Emotional Longitudinal Changes in PD Patients

In [32], two BMT groups of patients (only on medication) were tested. The first one, less advanced, was tested three times every half year (visit 1, visit 2, visit 3). In the second BMT group, more advanced patients were tested only once. All tests were performed with the following condition attributes: PDQ39, Epworth, depression score (Beck test), TMT A and B, disease duration, and fast EM. The decision attribute was UPDRS. With the help of Rough Set theory (RSES), rules describing the more advanced BMT group were constructed and used to predict disease progression over three visits in the less advanced BMT group of patients. Using all condition attributes, general rules gave accuracies as follows: visit 1—0.68, visit 2—0.86, and visit 3—0.88. When rules were related only to motor attributes, the accuracies were as follows: visits 1—0.80, 2—0.93, and 3—1.0. For rules related to cognitive attributes, the results were as follows: visit 1—0.50, visit 2—0.60, and visit 3—0.64. The higher accuracy can be interpreted as more similar patient symptoms. General and motor-related accuracies increased with disease progression (visit numbers), which means that the less advanced group of patients became more like the advanced group. However, this was not the case for cognition-related symptoms that gave lower accuracies, which means that their progressions were not as strongly correlated with disease development.
The influence of the patient’s emotions on the accuracy of the predictions of disease progression in the same group or different groups of patients was also tested through the depression score (Beck test) [33]. The progressions of the BMT group (only on medication) for visits 2 and 3 and the DBS group (deep brain stimulation) for visit 1 were compared based on the BMT symptoms during visit 1. The predictions were performed with the help of RSES and with standard neurological testing and EM parameters. Based on rules from first visit BMT patients, the prediction of symptoms (UPDRS) of BMT for visits 2 and 3 had accuracies of 0.7 and 0.7, but by adding the depression score, accuracies increased to 0.77 and 0.80 [33]. Similar predictions were calculated for the DBS group progression based on first visit BMT rules. Accuracies obtained for the DBS group were as follows: visit 1—0.64, visit 2—0.77, and visit 3—0.74. Adding the depression score to all attributes, improved accuracies of visit 1 to 0.77, visit 2 to 0.85, and visit 3 to 0.8 were demonstrated [33]. In summary, the depression score has a significant influence on predicting Parkinson’s disease progression.

10. EM in AD vs. PD

The impairment of the oculomotor system in AD manifested with longer RS latency along with higher variability in accuracy and speed [34]. Yang et al., 2012, found similarities between three groups: AD patients, patients with amnestic mild cognitive impairment (aMCI), and healthy elderly subjects [34]. All groups showed shorter latencies in the gap tests (when there is a time delay between the disappearance of the fixation spot and the appearance of the light spot in the periphery) than in the overlap tests (when the above spots’ appearance overlaps in time). However, in both tests, AD patients showed abnormally long saccade latencies. Although there was no significant difference in the accuracy (gain) and the velocity (both mean and peak velocity) between the three groups of subjects, AD patients showed an abnormally high coefficient of variation in the latency, accuracy, and speed of the reflexive saccades. There was a significant correlation between scores for the Mini-Mental State Examination (MMSE) and latencies of the saccades when comparing the MCI subjects to healthy elderly subjects [34].
Wilcockson et al. [35] explored AS eye movements in patients with amnestic and non-amnestic variants of MCI. There were 68 patients with dementia due to AD, 42 had amnestic MCI (aMCI), 47 had non-amnestic MCI (naMCI), and 92 were age-matched healthy controls (HC). The latencies for AS correction in the AD group were significantly longer than those for the HC and naMCI groups, but AS latencies in the AD group did not differ significantly from latencies in the aMCI group, even after age difference corrections [35]. They obtained similar results for the percentage of uncorrected AS errors. The AD and aMCI groups had similar and higher error rates than the naMCI and HC groups. This demonstrated that MCI patients are more likely to develop dementia due to AD than age-matched healthy adults. People with aMCI are at the highest risk of progressing to AD [36], and AS measurements might be an additional prognostic tool for predicting which people with MCI are more likely to progress to AD. It is worth noting that AS latency is a sensitive measure of the inhibitory process and is related to disease progression in the early stages of AD and PD.
In research by Pereira et al. [37], MCI sufferers were similarly impaired in their voluntary saccadic reaction times compared to AD sufferers, with a longer time to correct erroneous saccades.
Boxer et al. [38] compared saccade and antisaccade parameters in patients with frontotemporal dementia (FTD), patients with AD, and healthy subjects. The patients with AD showed an increased saccade latency compared to the FTD group during the horizontal saccade tasks. This might be related to the different dorsal parietal lobe roles in these two groups of patients [38]. In the AS task, all FTD and AD patients were impaired relative to the healthy subjects. The AD patients made fewer correct AS than controls, and they had more difficulty correcting saccade direction when they began from saccade instead of AS [37].
The relationships between AS parameters and measures of inhibitory control, attention, working memory, and self-monitoring showed correlations and common patterns reflecting deficits in executive function, confirming cognitive impairment in MCI and AD patients [37].


  1. Foley, J.A.; Lancaster, C.; Poznyak, E.; Borejko, O.; Niven, E.; Foltynie, T.; Abrahams, S.; Cipolotti, L. Impairment in Theory of Mind in Parkinson’s Disease Is Explained by Deficits in Inhibition. Park. Dis. 2019, 2019, 5480913.
  2. Przybyszewski, A.W. Theory of mind helps to predict neurodegenerative processes in Parkinson’s disease In Proceedings of the International Conference on Computational Science. Krakow, Poland, 16–18 June 2021; Springer: Berlin/Heidelberg, Germany, 2021; pp. 542–555.
  3. Briand, K.A.; Strallow, D.; Hening, W.; Poizner, H.; Sereno, A.B. Control of voluntary and reflexive saccades in Parkinson’s disease. Exp. Brain Res. 1999, 129, 38–48.
  4. Chambers, J.M.; Prescott, T.J. Response times for visually guided saccades in persons with Parkinson’s disease: A meta-analytic review. Neuropsychologia 2010, 48, 887–899.
  5. Przybyszewski, A.W.; Kon, M.; Szlufik, S.; Szymanski, A.; Habela, P.; Koziorowski, D.M. Multimodal Learning and Intelligent Prediction of Symptom Development in Individual Parkinson’s Patients. Sensors 2016, 16, 1498.
  6. Turner, T.H.; Renfroe, J.B.; Duppstadt-Delambo, A.; Hinson, V.K. Validation of a Behavioral Approach for Measuring Saccades in Parkinson’s Disease. J. Mot. Behav. 2017, 49, 657–667.
  7. Stuart, S.; Lawson, R.A.; Yarnall, A.J.; Nell, J.; Alcock, L.; Duncan, G.W.; Khoo, T.K.; Barker, R.; Rochester, L.; Burn, D.J.; et al. Pro-Saccades Predict Cognitive Decline in Parkinson’s Disease: ICICLE-PD. Mov. Disord. 2019, 34, 1690–1698.
  8. Perneczky, R.; Ghosh, B.; Hughes, L.; Carpenter, R.; Barker, R.; Rowe, J. Saccadic latency in Parkinson’s disease correlates with executive function and brain atrophy, but not motor severity. Neurobiol. Dis. 2011, 43, 79–85.
  9. Antoniades, C.A.; Xu, Z.; Carpenter, R.; Barker, R. The relationship between abnormalities of saccadic and manual response times in parkin- son’s disease. J. Park. Dis. 2013, 3, 557–563.
  10. Everling, S.; Fischer, B. The antisaccade: A review of basic research and clinical studies. Neuropsychologia 1998, 36, 885–899.
  11. Sledzianowski, A.; Szymanski, A.; Drabik, A.; Szlufik, S.; Koziorowski, D.M.; Przybyszewski, A.W. Measurements of antisaccades parameters can improve the prediction of Parkinson’s disease progression. In Proceedings of the Asian Conference on Intelligent Information and Database Systems, Yogyakarta, Indonesia, 8–11 April 2019; Springer: Berlin/Heidelberg, Germany, 2019; pp. 602–614.
  12. Waldthaler, J.; Stock, L.; Student, J.; Sommerkorn, J.; Dowiasch, S.; Timmermann, L. Antisaccades in Parkinson’s Disease: A Meta-Analysis. Neuropsychol. Rev. 2021, 31, 628–642.
  13. Waldthaler, J.; Stock, L.; Krüger-Zechlin, C.; Timmermann, L. Age at Parkinson’s disease onset modulates the effect of levodopa on response inhibition: Support for the dopamine overdose hypothesis from the antisaccade task. Neuropsychologia 2021, 163, 108082.
  14. Waldthaler, J.; Stock, L.; Sommerkorn, J.; Krüger-Zechlin, C.; Timmermann, L. Antisaccade Latency Is Sensitive to Longitudinal Change of Motor and Cognitive Symptoms in Parkinson’s Disease. Mov. Disord. 2020, 36, 266–268.
  15. Śledzianowski, A.; Szymanski, A.; Drabik, A.; Szlufik, S.; Koziorowski, D.; Przybyszewski, A.W. Combining results of different oculometric tests improved prediction of Parkinson’s disease development. In Proceedings of the Asian Conference on Intelligent Information and Database Systems, Phuket, Thailand, 23–26 March 2020; Springer: Berlin/Heidelberg, Germany, 2020; pp. 517–526.
  16. Antoniades, C.A.; Demeyere, N.; Kennard, C.; Humphreys, G.W.; Hu, M.T. Antisaccades and executive dysfunction in early drug-naive Parkinson’s disease: The discovery study. Mov. Disord. 2015, 30, 843–847.
  17. Fooken, J.; Patel, P.; Jones, C.B.; McKeown, M.J.; Spering, M. Preservation of Eye Movements in Parkinson’s Disease Is Stimulus- and Task-Specific. J. Neurosci. 2021, 42, 487–499.
  18. Koçoğlu, K.; Akdal, G.; Çolakoğlu, B.D.; Çakmur, R.; Sharma, J.C.; Ezard, G.; Hermens, F.; Hodgson, T.L. The effect of directional social cues on saccadic eye movements in Parkinson’s disease. Exp. Brain Res. 2021, 239, 2063–2075.
  19. Munoz, M.J.; Goelz, L.C.; Pal, G.D.; Karl, J.A.; Metman, L.V.; Sani, S.; Rosenow, J.M.; Ciolino, J.D.; Kurani, A.S.; Corcos, D.M.; et al. Increased Subthalamic Nucleus Deep Brain Stimulation Amplitude Impairs Inhibitory Control of Eye Movements in Parkinson’s Disease. Neuromodul. Technol. Neural Interface 2022, 25, 866–876.
  20. Przybyszewski, A.W.; Szlufik, S.; Dutkiewicz, J.; Habela, P.; Koziorowski, D.M. Machine learning on the video basis of slow pursuit eye movements can predict symptom development in Parkinson’s patients. In Proceedings of the Asian Conference on Intelligent Information and Database Systems, Bali, Indonesia, 23–25 March 2015; Springer: Berlin/Heidelberg, Germany, 2015; pp. 268–276.
  21. Śledzianowski, A.; Szymański, A.; Szlufik, S.; Koziorowski, D. Rough set data mining algorithms and pursuit eye movement measurements help to predict symptom development in Parkinson’s disease. In Proceedings of the Asian Conference on Intelligent Information and Database Systems, Dong Hoi City, Vietnam, 19–21 March 2018; Springer: Berlin/Heidelberg, Germany, 2018; pp. 428–435.
  22. Frei, K. Abnormalities of smooth pursuit in Parkinson’s disease: A systematic review. Clin. Park. Relat. Disord. 2020, 4, 100085.
  23. MacAskill, M.R.; Graham, C.F.; Pitcher, T.L.; Myall, D.J.; Livingston, L.; van Stockum, S.; Dalrymple-Alford, J.C.; Anderson, T.J. The influence of motor and cognitive impairment upon visually-guided saccades in Parkinson’s disease. Neuropsychologia 2012, 50, 3338–3347.
  24. Farashi, S. Analysis of vertical eye movements in Parkinson’s disease and its potential for diagnosis. Appl. Intell. 2021, 51, 8260–8270.
  25. Tabashum, T.; Zaffer, A.; Yousefzai, R.; Colletta, K.; Jost, M.B.; Park, Y.; Chawla, J.; Gaynes, B.; Albert, M.V.; Xiao, T. Detection of Parkinson’s Disease Through Automated Pupil Tracking of the Post-illumination Pupillary Response. Front. Med. 2021, 8, 645293.
  26. Tsitsi, P.; Benfatto, M.N.; Seimyr, G.; Larsson, O.; Svenningsson, P.; Markaki, I. Fixation Duration and Pupil Size as Diagnostic Tools in Parkinson’s Disease. J. Park. Dis. 2021, 11, 865–875.
  27. Bonnet, C.T.; Delval, A.; Singh, T.; Defebvre, L. Parkinson’s disease-related changes in the behavioral synergy between eye movements and postural movements. Eur. J. Neurosci. 2021, 54, 5161–5172.
  28. Zhang, J.; Zhang, B.; Ren, Q.; Zhong, Q.; Li, Y.; Liu, G.; Ma, X.; Zhao, C. Eye movement especially vertical oculomotor impairment as an aid to assess Parkinson’s disease. Neurol. Sci. 2020, 42, 2337–2345.
  29. Perkins, J.E.; Janzen, A.; Bernhard, F.P.; Wilhelm, K.; Brien, D.C.; Huang, J.; Coe, B.C.; Vadasz, D.; Mayer, G.; Munoz, D.P.; et al. Saccade, Pupil, and Blink Responses in Rapid Eye Movement Sleep Behavior Disorder. Mov. Disord. 2021, 36, 1720–1726.
  30. Chudzik, A.; Szymański, A.; Nowacki, J.P.; Przybyszewski, A.W. Eye-tracking and machine learning significance in Parkinson’s disease symptoms prediction. In Proceedings of the Asian Conference on Intelligent Information and Database Systems, Phuket, Thailand, 23–26 March 2020; Springer: Berlin/Heidelberg, Germany, 2020; pp. 537–547.
  31. Przybyszewski, A.W.; Chudzik, A.; Szlufik, S.; Habela, P.; Koziorowski, D.M. Comparison of Different Data Mining Methods to Determine Disease Progression in Dissimilar Groups of Parkinson’s Patients. Fundam. Informaticae 2020, 176, 167–181.
  32. Przybyszewski, A.W.; Nowacki, J.P.; Drabik, A.; Szlufik, S.; Koziorowski, D.M. IGrC: Cognitive and motor changes during symptoms development in Parkinson’s disease patients. In Proceedings of the Asian Conference on Intelligent Information and Database Systems, Phuket, Thailand, 23–26 March 2020; Springer: Berlin/Heidelberg, Germany, 2020; pp. 548–559.
  33. Przybyszewski, A.W.; Nowacki, J.P.; Drabik, A.; Szlufik, S.; Habela, P.; Koziorowski, D.M. Granular computing (GC) demonstrates interactions between depression and symptoms development in Parkinson’s disease patients. In Proceedings of the Asian Conference on Intelligent Information and Database Systems, Yogyakarta, Indonesia, 8–11 April 2019; Springer: Berlin/Heidelberg, Germany, 2019; pp. 591–601.
  34. Yang, Q.; Wang, T.; Su, N.; Xiao, S.; Kapoula, Z. Specific saccade deficits in patients with Alzheimer’s disease at mild to moderate stage and in patients with amnestic mild cognitive impairment. Age 2012, 35, 1287–1298.
  35. Wilcockson, T.D.; Mardanbegi, D.; Xia, B.; Taylor, S.; Sawyer, P.; Gellersen, H.W.; Leroi, I.; Killick, R.; Crawford, T.J. Abnormalities of saccadic eye movements in dementia due to Alzheimer’s disease and mild cognitive impairment. Aging 2019, 11, 5389–5398.
  36. Petersen, R.C.; Doody, R.; Kurz, A.; Mohs, R.; Morris, J.; Rabins, P.; Ritchie, K.; Rossor, M.; Thal, L.; Winblad, B. Current concepts in mild cognitive impairment. Arch. Neurol. 2001, 58, 1985–1992.
  37. Pereira, M.L.G.F.; Villa, M.; Koh, D.; Camargo, M.Z.A.; Belan, A.; Radanovic, M.; Pomplun, M.; Forlenza, O. Saccadic eye movements associated with executive function decline in mild cognitive impairment and Alzheimer’s disease: Biomarkers (non- neuroimaging)/novel biomarkers. Alzheimer’s Dement. 2020, 16, e040036.
  38. Boxer, A.L.; Garbutt, S.; Seeley, W.; Jafari, A.; Heuer, H.; Mirsky, J.; Hellmuth, J.; Trojanowski, J.; Huang, E.; DeArmond, S.; et al. Saccade abnormalities in autopsy-confirmed frontotemporal lobar degeneration and Alzheimer disease. Arch. Neurol. 2012, 69, 509–517.
Contributors MDPI registered users' name will be linked to their SciProfiles pages. To register with us, please refer to : , , , ,
View Times: 353
Entry Collection: Neurodegeneration
Revisions: 2 times (View History)
Update Date: 22 Feb 2023
Video Production Service