Visual Imagery and Scanpath Eye Movements while looking at Pictures: Normal subjects and patients with visual field defects

Wolfgang H. Zangemeister

Professor of Neurology, MD

Department of Neurology, University of Hamburg, Germany


In this study we report some findings about visual imagery in patients with stable homonymous hemianopia compared to healthy control subjects. These findings were obtained by analyzing the gaze control through recording of eye movements in different phases of viewing and imagery.

We used visual stimuli of different gradual complexity for the consecutive viewing and imagery phases.. With infrared oculography we recorded eye movements during this presentation phase and in three subsequent imagery phases in absence of the stimulus.

Analyzing the basic parameters of the gaze sequences (known as "Scanpaths"), we discovered distinct characteristics of the "Viewing–Scanpaths" and the "Imagery–Scanpaths" in both groups which suggests a reduced extent of the image within the cognitive representation. Furthermore we applied different similarity measures and found that the gaze sequences of the picture exploration phase compared to those of the imagery phases exhibited less similarity (whereas it still was significantly higher than the similarity values calculated for random strings) than the scanpaths of the imageriy series with each other.

This result could be reproduced both in hemianopic patients and normal subjects and we referred to it as "convergence of imagery". This finding suggests a strong top–down component in picture exploration: In both groups, healthy subjects and hemianopic patients, a mental model of the viewed picture evolves very soon, which substantially determines the eye movements. As our hemianopic patients showed analogous results to the normal subjects we conclude that these well adapted patients have a preserved cognitive representation despite their perceptual defect, which follows the same top–down vision strategies in the process of Visual Imagery.



Subjects and Apparatus

The patient collective consisted of 14 subjects from our university hospital with predominant homonymous hemianopia. The age range was 20-77 (Median: 53 years), the etiology of the sensory deficit was heterogeneous (mostly stroke, trauma, operation), as was the degree of adaptation to the deficit (varying from 5 days to 16 years, Median: 8 months). From the 8 complete and 6 incomplete hemianopias, perimeter showed left hemifield blindness in 7 patients, right hemifield blindness in 6 patients and one with a bitemporal deficit. All patients showed "Foveal sparing" in perimeter, no relevant oculomotor or cognitive dysfunctions and no Neglect syndromes either. In the control group, 20 normal subjects participated in our study. They neither showed current nor preceding neurological disorders and their age ranged from 19-58, Median: 26 years.

We used a 21" monitor and a head fixation device for the presentation of stimuli to the probands. Under standardized conditions the eye movements were recorded with infrared-reflection-oculography (ASL 210). The sampling rate was 200 Hz.

Experimental Protocol

Each proband was measured during a standardized routine that consisted of one stimulus presentation phase and 3 subsequent imagery phases. We used 6 different stimuli of graduated complexity which were all bordered by a reference frame. The subjects first had 10 seconds for viewing the presented picture according to varying instructions. In the following three separate imagery phases the probands then were asked to recall the pictures, just looking at the blank reference frame on the monitor and scan it again freely from the imagination (Figure 5 & 6).

The first 2 presentations (picture 1 served as "training" picture for acquiring of some routine with the task and was excluded from analysis) were used to measure the so called "searchpath", containing the task of searching objects within a picture. The next 4 pictures showed abstract, polyvalent and realistic stimuli. During this phase different scanpath tasks of increasing cognitive complexity were performed (s. Table 1).



Scanpath and Feature Ring Hypothesis of Vision

Although recent studies showed remarkable progress in enlightening different aspects of structural and functional compounds of "High–Level–Vision" processes there still remain various questions about the visual process as a whole. One fundamental questions is, whether the object recognition process is mostly parallel or serial and what kind of internal representation is used by the brain selecting and integrating the numerous complex and rapidly changing visual data that it is permanently exposed to.

The serial model suggests that the internal representation consists of single components and features which are sequentially "matched" to the object seen in the recognition phase. Because viewing consists of a sequence of alternating saccades and fixations, its measurable correlate can be described as the so called "scanpath". This term was introduced by Noton and Stark (1971). Their theory based upon the model that top–down processes associate single features of a scene on a higher cognitive level in a step–by–step manner. Within their studies they found marked similarities in recorded intrasubjective scanpaths of their probands when repetitively viewing the same picture. When viewing an object for an extended time the scanpath was repeated. From this Noton and Stark concluded the "feature ring hypothesis", which claims that the internal representation of a seen object is given by its cardinal sensory features and the motor traces (saccades) which connect these features. So recognition of objects works through a step–by–step scanning which corresponds to the actual "Feature Ring" (Figure 1). Based upon this model our approach to obtain information on the visual system was to register the eye movements of our probands during a visual task and then to analyze the resulting scanpaths.

Visual Imagery

The basic assumption of common anatomic structures serving the functions of perception ("vision") and Visual Imagery ("Imagery") in High-Level-Vision was confirmed in recent structural and imaging studies.Similar evidence for this was found by PET (Kosslyn et al., 1993, 1994; Roland et al., 1994), functional MRI (LeBihan et al., 1993) and regional blood flow recording (Goldenberg et al., 1989) studies localizing these functions in parieto- and temporo-occipital areas of the human cortex.

On the basis of these apparative results, Kosslyn et al. (1994) and Farah et al. (1994) developed an integrated model of active vision and visual imagery. This model basically postulates an interactive exchange of information between various cortical areas of different complexity. According to this theory the process of imagery works very similar to the recognition of previously seen pictures (matching them with the mental prototype) through projection of the visual information into a common medium. This projection of an extrastriate "High-Level" area into the retinotopical organized V1-area (which serves as "window" for incoming retinal images as well) finally corresponds to the visualization of the constructed image and represents the visual working memory. This means, that the striate cortex plays the role of a visual "buffer" offering incoming data to an "attention window" (Treisman et al., 1980) that selectively focuses on the interesting features of the whole picture. Like in the process of analyzing external images, different subsystems are involved to define these objects concerning their major aspects like shape, color, localization and size (Kosslyn, 1994). There is a direct link between the lateral geniculate nucleus (LGN) and V1 in the sulcus calcarinus of the occipital cortex given through the radiatio optica. Further connections exist from LGN to the extrastriate cortex in area 18 and 19. Zeki (1976, 1978, 1983 and 1993) was the first to show an association of the areas V2-5 in the integration of different modalities of visual information like color, shape and motion in the rhesus monkey.

In the literature few works can be found which approach the analyzation of eye movements during visual imagery. In 1968 Hebb asked in "Concerning Imagery", whether eyes scan a visualized scene the same way they do in reality. He concluded that eye movements have to play an essential, organisatory role in completing the fragments of a whole picture.

Our paradigm of comparing "Viewing"-Scanpaths with "Imagery"-Scanpaths was first carried out by Brandt et al.(1997), using just one kind of abstract stimulus and a small number of healthy subjects. They were able to show that after presentation of a stimulus to the probands eye movements occur in the subsequent imagery phase. Furthermore, using string editing methods in the scanpath analysis, they found that these scanpaths were not random but similar to the viewing phase.

Homonymous hemianopia and rehabilitative aspects

Hemianopia, defined as a complete visual field defect is divided into different forms according to the site of the lesion. The predominantly occurring homonymous form usually shows macular sparing. Common etiologies of this disorder is the ischemic cerebrovascular event followed by bleedings and trauma. Often hemianopia is associated with other cognitive dysfunctions like aphasia and visual hemineglect. Rossi et al. (1990) found that more than 20 % of patients with stroke treated in rehabilitation centers expose hemianopic symptoms. The impact of this sensory deficit depends on size and localization of the lesion, impairing patients in visual information processing in many ways. Hemianopia usually causes problems exploring the blind hemifield causing patients to perform hypometric, slow down low amplitude saccades and handicaps them more or less severely in orientation and safety in everyday living. Prospective studies of the natural course of vascular retrogenicular visual field defects showed that spontaneous restitution (e.g. axon-sprouting) in the blind hemifield takes place within the first 6 months after the event and that the average visual field gain is about 16 % (Hier et al., 1983; Messing et al., 1986) in perimeter. To some degree oculomotor training strategies can compensate the sensory deficit (Riddoch, 1917; Poppelreuter, 1917; Gassel et al., 1963; Meienberg et al., 1981 & 1983; Zihl et al., 1981 & 1988; Zangemeister et al., 1982 & 1986; Schöpf et al., 1993). Pommerenke et al. (1989) found that a specific systematic exploration practice through perimetric saccade training improves visio-spatial orientation in these patients. Furthermore, Zangemeister et al. (1999) investigated the influence of cognitive motor gaze control strategies on the rehabilitation of visual field defects in hemianopic patients and found significant improvement in their visual behavior after taking part in a special cognitive training of gaze control.

A study of Butter et al. (1997) tested visual imagery in hemianopic patients with occipital lesions using special imagery and perceptual control tasks. On the basis of their results, they postulated an impaired visual imagery in these patients because compared to a control group they performed worse in the imagery task when perceiving the stimulus ipsilateral to their visual field defect. They concluded that this findig supports the view that visual imagery involves topographically organized visual areas of the occipital lobe.

In our study we primarily tried to obtain information about the consistency and reproducibility of internal visual image representations in hemianopic patients in the context of active "high-level-vision". Therefore the question was whether, in comparison with healthy subjects this mental image can be thought of as being intact despite the visuocortical sensory deficit.