Eye movement function captured via an electronic tablet informs on cognition and disease severity in Parkinson's disease

Abstract

Studying the oculomotor system provides a unique window to assess brain health and function in various clinical populations. Although the use of detailed oculomotor parameters in clinical research has been limited due to the scalability of the required equipment, the development of novel tablet-based technologies has created opportunities for fast, easy, cost-effective, and reliable eye tracking. Oculomotor measures captured via a mobile tablet-based technology have previously been shown to reliably discriminate between Parkinson's Disease (PD) patients and healthy controls. Here we further investigate the use of oculomotor measures from tablet-based eye-tracking to inform on various cognitive abilities and disease severity in PD patients. When combined using partial least square regression, the extracted oculomotor parameters can explain up to 71% of the variance in cognitive test scores (e.g. Trail Making Test). Moreover, using a receiver operating characteristics (ROC) analysis we show that eye-tracking parameters can be used in a support vector classifier to discriminate between individuals with mild PD from those with moderate PD (based on UPDRS cut-off scores) with an accuracy of 90%. Taken together, our findings highlight the potential usefulness of mobile tablet-based technology to rapidly scale eye-tracking use and usefulness in both research and clinical settings by informing on disease stage and cognitive outcomes.

Figure 1

Distribution of participant age (a) and participants’ clinical scores: UPDRS-III (b), UPDRS (c), H&Y (d), MoCA (e), TMTA (f), TMTB (g), HVLT (h), and COWAT-CFL (i). Spearman correlations between participants' clinical scores (j).

Figure 2

Correlations between eye-tracking parameters and clinical scores. Features with Spearman rho correlation coefficient greater than 0.33 for one clinical outcome measure are shown. Spearman rho correlation coefficient below this threshold of 0.33 are not shown. For a complete correlation matrix see Table 2 and Supplementary Fig. 3. Eye-tracking parameter label colors indicate different oculomotor tasks. PS: pro-saccade, SP: smooth pursuit, OKN: Optokinetic nystagmus.

Figure 3

Correlations between select eye-tracking parameters and clinical scores UPDRS-III (a), UPDRS (b), H&Y (c), MoCA (d), TMTA (e), TMTB (f), HVLT (g), and COWAT-CFL (h). All Spearman's rho correlation values were calculated using the raw data. Linear regressions with 95% confidence intervals are shown for visualization purposes only. PS: pro-saccade, SP: smooth pursuit.

Figure 4

(a-h) Scatterplots of the relationship between the study participants’ clinical scores and the corresponding predicted value obtained by partial least squares regression analysis using the oculomotor parameters as predictors. (i) Heatmap visualization of the relative contribution (normalized absolute value of standardized regression coefficients) of each oculomotor parameter to each partial least squares regression predictor. Dark squares indicate lesser contributions to the model whereas lighter/yellow squares indicate greater contributions. Absent squares indicate that the parameter was not used in the final model. Eye-tracking parameter label colors indicate different oculomotor tasks. PS: pro-saccade, SP: smooth pursuit, OKN: Optokinetic nystagmus.

Figure 5

Performance of the support vector classifier. (a) Mean ROC curve for the logistic regression classifier across fivefold cross validation. Grey lines represent each fold. (b) Confusion matrix for classification of mild (UPDRS-III 0–35) and moderate (UPDRS-III 36–57) Parkinson's Disease using eye tracking parameters. (c) Confusion matrix for classification of mild (UPDRS-III 0–35), moderate (UPDRS-III 36–57) and severe (UPDRS-III > 57) Parkinson's Disease using eye tracking parameters.