Abstract

P. Tseng, I. G. M. Cameron, D. P. Munoz, L. Itti, Differentiating Patients from Controls Based on Correlation between Salience and Gaze, In: Collaborative Research in Computational Neuroscience Annual Meeting, Los Angeles, California, Jun 2008. (Cited by 2)

Abstract: Several studies have shown that eye movements and certain complex visual functions are influenced by diseases such as Attention Deficit Hyperactivity Disorder (ADHD), Fetal Alcohol Spectrum Disorders (FASD) and Parkinson's Disease (PD). Here we examine how bottom-up (stimulus-driven) attentional selection mechanisms may differ between patient and control populations, and we take advantage of the difference to develop classifiers to differentiate patients from controls. We tracked gaze of five groups of observers (15 control children, aged 7-14; 6 ADHD children, aged 9-15; 4 FASD children, aged 9-15; 12 control elderly, aged 66-79; and 9 PD elderly, aged 53-68) while they freely viewed MTV-style videos. These stimuli are composed of short (2-4 seconds), unrelated clips of natural scenes to reduce top-down (contextual) expectations and emphasize bottom-up influences on gaze allocations at the scene change. We used a saliency model to compute bottom-up saliency maps for every video frame. Saliency maps can be computed from a full set of features (color, intensity, orientation, flicker, motion) or from individual features. Support-vector-machine classifiers were built for each feature contributing the saliency map and for the combination of them. Leave-one-out was used to train and test the classifiers. Two classification experiments were performed: (1) between ADHD, FASD and control children; (2) between PD and control elderly. Saliency maps computed with all features can well differentiate patients and control populations (correctness: experiment 1 - 80%; experiment 2 - 95.24%). Additionally, saliency maps computed from any one feature performed nearly as well (correctness: experiment 1 - 92% for flicker; experiment 2 - 100% for color and flicker). This study demonstrates that the bottom-up mechanism is greatly influenced by ADHD, FASD and PD, and the difference can serve as a probable diagnosis/screening tool for clinical applications.

Themes: Model of Bottom-Up Saliency-Based Visual Attention, Model of Top-Down Attentional Modulation, Computational Modeling, Human Psychophysics, Medical Research