The Effects of Normal Aging, Subjective Cognitive Decline, Mild Cognitive Impairment, or Alzheimer's Disease on Visual Search

Background (cid:0) To examine the progressively developed visual-search deciency associated with Alzheimer's disease (AD). Methods: Healthy-younger adults (34), healthy-older adults (normal-aging control, NC, 30), adults with subjective cognitive decline (SCD, 14), amnestic mild cognitive impairment (aMCI, 20), or mild AD (15) participated in this study. To determine whether 1 of 4 letters presented at 4 symmetrically-located positions differed from the other 3, when the 4 letters were masked by either other randomly positioned and oriented letters or random-pixel noise. Meanwhile, eye movements were tracked. Results: In all the participants, with the stimulus-presentation time being longer, the visual-search performance improved, and both the eye interest-area rst xation duration (IFFD) and the interest-area-xation count (IFC) increased. Particularly under the noise-masking condition, the AD group performed the worst at stimulus-presentation times between 300 and 900 ms. The aMCI group, but not the SCD group, performed worse than the NC group at the stimulus-presentation time of either 300 or 500 ms. The IFC was higher in all the patient groups than that in the NC group, and distinguishable between participants with AD and those with SCD or aMCI. Conclusions: The visual-search performance combined with eye-moment tracking under the noise-masking condition can be used for distinguishing AD from normal aging, SCD, and aMCI. xation counts (IFC) in the younger-healthy group and that in the older-healthy group under the LM condition. 2 6 ANOVA between age time signicant [F (5,168) = 8.411, p < 0.001, ηp 2 = 0.72]. Post-tests showed that at the longer times (300, 500, 700, 900 and 1100 ms), the IFC difference between the younger group and the older group under the LM condition was signicant (for all the stimulus presentation times adjusted p < 0.001).


Procedures
The participants were given a short experimental training as formal experiment (usually continuous and complete trials for 8 times), through the simulation practice of the experiment, until the participants clearly understood the operational requirements of the experiment.
In the formal experiment, the 3 or 4 identical target letters were named "reference (ordinary)" letters, and the different (odd) one was named "different (odd) letter" (Figure 2). Upon a trial started, a "+" viewpoint was presented for 500 ms, allowing the participant to pay attention to the following stimuli. The "+" viewpoint was followed by a presentation time of target letters with 100, 300, 500, 700, 900, or 1100 ms. After the stimulus presentation, a masking scatter with Gaussian distribution was presented immediately for 100 ms, and the visual afterimage was eliminated. Finally, a blank screen with a duration of 3000 ms was presented. During the letter presentation, the participant watched the target letters against the masking background and made a judgment: pressing the keypad "1" when all the 4 target letters were the same, and pressing the keypad "2" when the 4 target letters were not the same (the reference letters were different from the odd one). In total 24 trials for each participant.

Data Analyses
Data View software was used to review data and draw the area of interest. Statistical analyses were performed using SPSS 20. GraphPad Prism 7 was used to map the results after data analyses. Two-way ANOVAs and post hoc tests were performed. Simple effects analyses were required if interactions were signi cant. The data were processed with GraphPad Prism 7.0 to plot the resulting images.
For statistical analyses, data were divided into behavioral data and eye-movement data. According to the categories of participants, participants were divided into the healthy control group and the patient groups, and the control group was further divided into the younger-adult group and the older-adult group. There were three patient groups including the SCD, AMCI and AD groups. The stimulus presentation time variable contained 6 levels, including 100, 300, 500, 700, 900, and 1100 ms. The masker variable contained two levels, including the letter informational masker and the noise energetic masker.
For the two healthy groups under each masker type, a 2 (age group: younger, older) × 6 (stimulus presentation time: 100, 300, 500, 700, 900, 1100 ms) ANOVA was rst conducted and then post hoc analyses were carried out to make comparisons between groups at each of the stimulus presentation times. For the four older-adult groups under each masker type, a 4 (group: NC, SCD, MCI, AD) × 6 (stimulus presentation time: 100, 300, 500, 700, 900, and 1100 ms) ANOVA was rst conducted and then post hoc analyses were carried out to make comparisons between groups at each of the stimulus presentation times."

Results
The Effects of Normal Aging Figure 3 shows the results of the visual-search behavioral performance (panels a,b) and the eye tracking (panels c,d,e,f) in the healthy-younger and the healthy-older groups under either the letter masking (LM) condition (left panels) or the random noise (RN) masking condition (right panels). The visual-search performance was poor in the two groups under the LM condition, indicating that the LM condition had a much stronger masking effect than the RN-masking condition.

Visual-search performance
Under the LM condition (Figure 3a), with the increase of the stimulus presentation time, the performance in the younger group but not in the older group improved. A 2 (age group: younger, older) × 6 (stimulus presentation time: 100, 300, 500, 700, 900, 1100 ms) ANOVA showed that the interaction between the stimulus presentation time and group was signi cant [F (5,168) = 21.83, p < 0.001, ηp 2 = 0.59]. Post hoc analyses showed that the differences in visual-search performance between the younger group and the older group under the LM condition was signi cant (the performance of the younger group was signi cantly better than that of the older group) only when the stimulus presentation time was either 700 ms (adjusted p = 0.002) or 1100 ms (adjusted p = 0.0394). Figure 3b shows the visual search behavioral performance in the heathy-younger group and the healthy-older group under the RNmasking condition. With the increase of the stimulus presentation time, the performance in both groups gradually improved, and the difference between the two groups was marked. A 2 × 6 ANOVA showed that the interaction of stimulus presentation time and the group was signi cant [F (5,168) = 37.47, p < 0.001, ηp 2 = 0.46]. Post hoc analyses showed that under the RN-masking condition, at each of the stimulus presentation times, the behavioral performance of the younger group was signi cantly better than that of the older group (for each of the 6 presentation times, adjusted p < 0.01).
Eye movements Figure 3c shows the interest-area rst xation duration (IFFD) in the younger group and that in the older group under the LM condition. A 2 × 6 ANOVA showed that the interaction between group and stimulus presentation time was signi cant [F (5,168) = 38.07, p < 0.001, ηp 2 = 0.78]. Post hoc analyses showed that the IFFD difference between the younger group and the older group under the LM condition was signi cant when the stimulus-presentation time was 300 ms (adjusted p < 0.001), 700 (adjusted p = 0.005), 900 (adjusted p < 0.001), or 1100 ms (adjusted p < 0.001). Figure 3d shows the IFFD in the younger healthy group and that in the older healthy group under the RN-masking condition. A 2 × 6 ANOVA showed that the interaction between age difference and stimulus presentation time was signi cant [F (5,168) = 30.64 (p < 0.001), ηp 2 = 0.38]. Post-tests showed that the IFFD difference between the younger group and the older group under the RNmasking condition was signi cant when the stimulus-presentation time was 300 (adjusted p = 0.015), 700 (adjusted p < 0.001), or 900 ms (adjusted p < 0.001; 1100 ms, adjusted p < 0.001). Figure 3e shows the maximum number of interest-area xation counts (IFC) in the younger-healthy group and that in the olderhealthy group under the LM condition. A 2 × 6 ANOVA showed that the interaction between age group and stimulus presentation time was signi cant [F (5,168) = 8.411, p < 0.001, ηp 2 = 0.72]. Post-tests showed that at the longer times (300, 500, 700, 900 and 1100 ms), the IFC difference between the younger group and the older group under the LM condition was signi cant (for all the 5 stimulus presentation times adjusted p < 0.001). Figure 3f shows the maximum number of IFC of the younger healthy group and that of the older healthy group under the RNmasking condition. A 2 × 6 ANOVA showed that the interaction between age group and stimulus presentation time was signi cant [F (5,168) = 5.637, p < 0.001, ηp 2 = 0.42]. Post hoc analyses showed that at the longer stimulus presentation times (300, 500, 700, 900, and 1100 ms), the IFC difference between the younger group and the older group under the RN-masking condition was signi cant (for all the ve stimulus presentation times, adjusted p < 0.001).

The Normal Relationship between the Visual-Search Performance and the Eye Movement
To demonstrate the normal relationship between the visual-search performance and the eye movement, Figure 4 shows the visualsearch behavioral performance as a function of the average number of IFC across individual participants in the younger healthy group under either the LM condition (left panel) or the RN-masking condition (right panel) at the 6 different stimulus presentation times. As mentioned above, both the visual-search performance improved and the IFC increased as the stimulus-presentation time became longer. There is a positive linear correlation between the behavioral performance and the IFC at either masking condition. The mathematical formula of behavioral performance (the ordinate) as a function of IFC (the abscissa) was also established Comparisons between the Older-Healthy Group and Patient Groups Figure 5 shows the visual-search behavioral performance and the eye-tracking results in the healthy-older (normal-control, NC) group and the 3 older-patient groups (SCD, MCI, and AD) under either the LM condition (left panels) or the RN-masking condition (right panels). The visual-search performance was also very poor in each group under the LM condition than the that under the RNmasking condition (Figure 5a,b).

Visual-search behavioral performance under the LM condition
For the visual-search performance under the LM condition (Figure 5a), a 6 (stimulus presentation time: 100, 300, 500, 700, 900, and 1100 ms) × 4 (group: NC, SCD, MCI, AD) ANOVA showed a signi cant interaction between stimulus presentation time and group [F (15,336) = 9.165 p < 0.001, ηp 2 = 0.30]. Post hoc tests showed that there was no signi cant difference in behavioral performance between the NC group and the SCD group, the aMCI group or the AD group at each of the stimulus presentation times (for all the 6 stimulus presentation times, adjusted p > 0.999). There was no signi cant difference in behavioral performance between the SCD group and the aMCI group or the AD group at each of the stimulus presentation times (for all the 6 stimulus presentation times, adjusted p > 0.999). There was no signi cant difference in behavioral performance between the aMCI group and the AD group at each of the stimulus presentation times (for all the 6 stimulus presentation times, adjusted p > 0.999).
Visual-search behavioral performance under the RN-masking condition The behavioral performance of the NC group was signi cantly better than that of the aMCI when the stimulus presentation time was either 300 ms (adjusted p = 0.001) or 500 ms (adjusted p = 0.02).
The behavioral performance of the NC group was signi cantly better than that of the AD group when stimulus presentation time was 300 ms (adjusted p = 0.01), 500 ms (adjusted p = 0.005), 700 ms (adjusted p = 0.013), or 900 ms (adjusted p < 0.001).
The behavioral performance of SCD was signi cantly better than that of the aMCI group only when the stimulus presentation time was 300 ms (adjusted p = 0.003).
The visual-search performance of the SCD group was signi cantly better than that of the AD group when the stimulus presentation time was 300 ms (adjusted p = 0.019), 700 ms (adjusted p = 0.047), or 900 ms (adjusted p < 0.001).
There was no signi cant difference in behavioral performance between the aMCI and AD groups at each of the stimulus presentation times (for all adjusted p > 0.060).
Eye movement IFFD under the LM condition Figure 5c shows the IFFD in these groups with older participants under the LM condition. With the increase of the stimulus presentation time, the IFFD in each of groups increased. A 6 × 4 ANOVA showed that there was a signi cant interaction between stimulus presentation time and group [F (15,336) = 3.221, p < 0.001, ηp 2 = 0.49]. Post hoc tests showed that under the LM condition, the IFFD of the SCD group was signi cantly longer than that of NC group when stimulus presentation time was 500 ms or larger (for the four stimulus presentation times, adjusted p < 0.001)..
The IFFD of the aMCI group was signi cantly longer than that of NC group when the presentation time was either 500 ms (adjusted p = 0.006), 900 (adjusted p < 0.001) or 1100 ms (adjusted p < 0.001).
The IFFD of the AD group was signi cantly longer than that of the NC group only when the stimulus presentation time was 1100 ms (adjusted p < 0.001).
The IFFD of the SCD group was signi cantly longer than that of the aMCI group only when the stimulus presentation time was 1100 ms (adjusted p = 0.002).
The IFFD of the SCD group was signi cantly longer than that of the AD group when stimulus presentation time was also 1100 ms (adjusted p = 0.034).
There was no signi cant difference in IFFD between the aMCI group and the AD group at each of the 6 stimulus presentation times (for all the 6 stimulus presentation times, adjusted p > 0.999).
Eye movement IFFD under the RN-masking condition Figure 5d shows the IFFD in the groups with older participants under the RN-masking condition. The IFFD became longer with the increase of the stimulus presentation time similarly for all the groups with older participants. A 6 × 4 ANOVA showed that under the RN-masking condition there was a signi cant interaction between stimulus presentation time and group [F (15,336) = 29.92, p < 0.001, ηp 2 = 0.32]. Multiple comparisons showed that when stimulus presentation time was 500 ms or above, the IFFD of the SCD group was signi cantly longer than that of the NC group (for the 4 stimulus presentation times, adjusted p4 < 0.001).
The IFFD of the aMCI group was signi cantly longer than that of the NC group when the stimulus presentation time was 500 ms and longer (for the 4 stimulus presentation times, adjusted p < 0.001).
The IFFD of the AD group was signi cantly longer than that of the NC group when stimulus presentation time was also 500 ms or longer (for the 4 stimulus presentation times, adjusted p < 0.001).
The IFFD of the aMCI group was signi cantly longer than that of SCD group only when stimulus time was 1100 ms (adjusted p = 0.024)..
The IFFD of the AD group was signi cantly longer than that of the SCD group only when the stimulus presentation time was either 900 ms or 1100 ms (for the 2 stimulus presentation times, adjusted p < 0.001).
There was no signi cant difference in IFFD between the aMCI and AD groups at each of the stimulus presentation times (all adjusted p > 0.999). The IFC of the aMCI group was signi cantly different from that of the NC group when the stimulus presentation time was 500 ms or longer (for each of the 4 stimulus presentation times, adjusted p < 0.001).

The Eye movement IFC under the LM condition
The IFC of the AD group was signi cantly higher than that of the NC group when the stimulus presentation time was either 300 ms or 900 ms (for the 2 stimulus presentation times, adjusted p < 0.001).
The IFC of the SCD group was signi cantly higher than that of the aMCI group when the stimulus presentation time was 500, 700, or 1100 ms (for the 3 stimulus presentation times, adjusted p < 0.001).
The IFC of the SCD group was signi cantly different from that of the AD group when the stimulus presentation time was 100, 500, 900, or 1100 ms (for the 4 stimulus presentation times, adjusted p < 0.001).
The IFC of the AD group was signi cantly higher than that of the aMCI group when the stimulus presentation time was 500 ms and above (for the 4 stimulus presentation times, adjusted p < 0.001).. The IFC of the aMCI group was signi cantly higher than that of the NC group only when the stimulus presentation time was 500 ms (for this stimulus presentation time, adjusted p < 0.001).. The IFC of the AD group was signi cantly higher than that of the NC group when the stimulus presentation time was 500, 700, or 900 ms (for the 3 stimulus presentation times, adjusted p < 0.001)..

The IFC under the RN-masking condition
The aMCI group was signi cantly different that of the SCD group when the stimulus presentation times was100 ms or 500 ms (adjusted p <0.001)].
The IFC of the AD group was signi cantly higher than that of the SCD group when the stimulus presentation time was 100, 700, or 900 ms (for the 3 stimulus presentation times, adjusted p < 0.001)..
The IFC of the AD group was signi cantly higher than that of the aMCI group when the stimulus presentation time was 500, 700, or 900 ms (for the 3 stimulus presentation times, adjusted p < 0.001).

Discussion
For the visual research performance in this study, the spatial pattern of the positions of the 4 target letters provided low-spatialfrequency priming information for facilitating the recognition of the target letters with high-spatial information. Thus, the feedback propagation mechanism driven by the low spatial frequency information was important for the visual search performance based on a coarse-to-ne integration of information. Studies using different stimuli have con rmed the existence of such a process: low spatial frequency information is rst processed and quickly projected from the primary visual cortical area to the higher-order cortical areas, and feedback information is then generated in the higher-order region to top-down modulate the processing of the high spatial frequency information in the lower-level regions [18][19][20][21][22][23] . This strategy is useful for solving the crowding problems for visual search.
The Barnikol et al. [24] study has shown that using magnetoencephalography (MEG) with high temporal resolution, the activation of the left orbitofrontal cortex caused by object recognition is 50 ms earlier than that activation in the related visual sensory areas.
And this earlier activation in the left orbital frontal cortex is directly affected by the low frequency information of visual stimuli, supporting the view that low-frequency information is rst processed in the orbitofrontal cortex to form a prediction of the input images and then transmitted back to the ventral pathway of the temporal lobe. The visual-search paradigm used in this study speci cally examine the integration of visual information from coarse (associated with the target-letter position globe attention) to ne (associated with the target-letter-feature local attention) information process in letter recognition.
As mentioned in the Introduction, AD is related with progressive impairments of cognitive functions, and the development of AD may pass a few stages along with aging, including SCD, aMCI, mild AD, moderate AD, and nally, severe AD. This study was to establish a cognitive paradigm that is useful for screening people with AD in an e cient and quick way.
In this study, there were two types of maskers: the informational letter masker containing randomly positioned and oriented different letters and the energetic random-pixel noise masker. The letter masker had strong masking impact and caused oor effects.

The Effects of Normal Aging
The results of this study showed that the visual-search behavioral performance in younger-healthy group and older-healthy group under the noise masking condition could be better distinguished than that under the LM condition, because the masking effect of the LM was too strong. Speci cally, under the RN-masking condition, the performance of the healthy-older group was signi cantly poorer than that of the healthy-younger group at each of the 6 stimulus presentation times. However, under the LM condition, the visual-search performance of the healthy-older group was signi cantly poorer than that of the healthy-younger group only when the stimulus presentation time was either 700 ms or 1000 ms. Thus, LM condition had much stronger masking effects than the RNmasking condition. The Yang et al. study [25] has con rmed that compared to the RN-masking condition, the LM condition has a greater interference effect on visual search and causes greater processing load in the primary visual cortex and the secondary visual cortex.
Moreover, the eye movements were also markedly different between the healthy-younger group and the healthy-older group under either the LM or the RN-masking condition. Speci cally, either the IFC or the IFFD under either the LM condition or the RN masking condition was signi cantly different between the 2 healthy groups when the stimulus presentation time was 300 ms or longer.
Particularly, the IFC under the RN masking condition in the healthy-older group was markedly higher than that of the healthyyounger group when the stimulus presentation time was no less than 300 ms. Also, in the healthy-younger group the IFC under the RN-masking condition was correlated with the visual search performance, showing the normally functional association between the visual-search performance and the eye movement.

Visual-Search Performance and Eye Movement
The results of this study showed that under the RN-masking condition, but not the LM condition, patients with SCD were not signi cantly different in the visual-search performance from the healthy-older participants. However, patients with aMCI performed in the visual search signi cantly worse at stimulus presentation times of either 300 or 500 ms, at which the visual-search performance in patients with aMCI was also signi cantly poorer than that in patients with SCD.
The stimulus presentation times, at which patients with AD were signi cantly worse in the visual-search performance than the healthy-older participants (NC) under the RN-masking condition, included 300, 500, 700, and 900 ms. The visual-search performance in patients with AD was worse than that in patients with SCD at the stimulus presentation times of 300, 700, and 900 ms. Moreover, there was no signi cant difference in the behavioral performance between aMCI and AD at each of the 6 stimulus presentation times. Thus, along the order of NC, SCD, aMCI, and AD, the visual-search performance successively got worse.
The stimulus presentation times, at which patients with AD were markedly different from the healthy-older participants (NC) in eye movement IFC under the LM condition only included 900 ms. The stimulus presentation times, at which patients with AD were markedly different from the NC participants in IFC under the RN-masking condition included 300 ms and 900 ms.
At the stimulus presentation time of 900 ms, patients with AD were also markedly different from both patients with SCD and patients with aMCI in IFC under either the LM condition or the RN-masking condition.

Conclusions
By combining the visual search task and eye-movement measurement, this study established a new paradigm for screening people with AD. Particularly under the RN-masking condition, when the stimulus presentation time is set at 900 ms, people with AD are markedly different from healthy older people in both the visual-search performance and eye-movement IFC. At these stimulus conditions, people with AD are also different from people with SCD and people with aMCI to certain degree in both the visualsearch performance and eye-movement IFC.

Availability of data and material
Data and materials of this study are available upon requests.

Competing interests
The authors declare that they have no competing interests.

Funding
This study was supported by a grant from the '973' National Basic Research Program of China (2015CB351800) (for the design of the study, the collection, analysis, and interpretation of data, and writing the manuscript), the National Natural Science Foundation of China (31771252, 81970996)(for the design of the study, the collection, analysis, and interpretation of data, and writing the manuscript), and the Tianjin Philosophy and Social Science Project (TJJX15-002)(for the design of the study, the collection, analysis, and interpretation of data, and writing the manuscript).
Authors' contributions CWX conducted the experiments and wrote and rst manuscript draft; YT designed the study and wrote the manuscript; CMW designed the study and wrote the manuscript; HBY designed the study and wrote the manuscript; LL designed the study and wrote the manuscript. All authors have read and approved the manuscript.