Identification of 2D Images in Cerebral Visual Impairment (CVI) Based on Features and Gaze Behavior

Cerebral visual impairment (CVI) is a pediatric brain-based visual disorder caused by perinatal neurological injury. Anecdotal evidence suggests that individuals with CVI have difficulties recognizing abstract (e.g. cartoon) compared to real (e.g. photograph) images of familiar objects. To investigate this observation, CVI participants (n=7; mean age=17.71 yrs; Wechsler verbal IQ=84/167) and neurotypical controls (n=7; mean age=22.86 yrs; Wechsler verbal IQ=127/167) were instructed to verbally identify a series of 2D images while gaze behavior was recorded with an eye tracker (Tobii 4C; 90 Hz). 60 test images were comprised of 12 different common objects, each with 5 possible image types (real image, realistic image with color, realistic black and white outline, abstract image with color, and abstract black and white) and presented in pseudorandom order. Subjects manually advanced the presentation after identifying the image aloud. Image identification accuracy and reaction time (RT) were recorded. Compared to controls, CVI participants were less accurate in identifying images (accuracycontrols=100%; accuracyCVI=97%) and were slower to respond (RTcontrols=0.95 sec ± 0.06 SD; RTCVI=1.42 sec ± 0.06 SD). For both CVI and control subjects, average reaction time was greater for less realistic images. In CVI, the most common errors and longest RTs were found for images without color, suggesting that color cues (rather than image outline form) may be most helpful for identification. A receiver operating curve (ROC) analysis comparing the distribution of eye gaze patterns to a Graph-Based Visual Saliency (GBVS) saliency model revealed that CVI subjects had lower mean ROC values (0.86 ± 0.02 SD) compared to controls (0.95 ± 0.01 SD). While real images were most easily recognized, they showed the greatest discrepancy between groups with respect to ROC scores (ROCcontrols=0.97 ± 0.01 SD; ROCCVI=0.84±0.2 SD). These findings may provide clues to help better understand which image cues contribute to identification difficulties in CVI.