Tablet-Based Assessment of Picture Naming in Prodromal Alzheimer’s Disease: An Accessible and Effective Tool for Distinguishing Mild Cognitive Impairment from Normal Aging Open Access

Alzheimer’s disease (AD) affects 57 million people worldwide. AD is the most common cause of dementia [1]. In the prodromal phase of AD, known as mild cognitive impairment (MCI), individuals have cognitive deficits but can still perform daily tasks [2]. While MCI can be caused by non-AD etiologies, approximately one-third of MCI patients become functionally impaired within 5 years, transitioning to the Alzheimer’s dementia phase [2‒4].

Subtle cognitive changes in the MCI stage make screening difficult, especially by non-specialists [5‒7]. Furthermore, diagnosis of MCI due to AD may rely on invasive and expensive biomarker tests like PET and spinal fluid analysis [8]. However, visual perception and eye movement changes may be detectable in the MCI stage, preceding AD [9‒13], offering an inroad to less-invasive and cost-effective screening measures.

Specifically, visual perception-based testing, like rapid automatized naming (RAN) and low-contrast letter acuity (LCLA), may be alternative tools for identifying MCI before progression to AD [12]. RAN tasks require quick reading of numbers or naming of objects. Number naming involves numerical cognition and oculomotor control, while picture naming involves visual processing and semantic memory [14, 15]. Slower picture naming is associated with AD [16]. For example, an early study demonstrated that the Boston Naming Test of picture naming effectively differentiates Alzheimer’s patients from controls [17]. Recent literature reinforces that the Boston Naming Task is also effective in distinguishing MCI from controls [18]. A newer picture naming task called the Multilingual Naming Test, can also distinguish individuals with MCI from cognitively normal individuals [19, 20]. Deficits in number naming RAN tasks are also linked to AD. For example, slow performance in the King-Devick Test, which combines assessments of goal-directed eye movements, saccades, and verbal number naming, is associated with AD and MCI [21, 22]. Other work also shows that oculomotor functions involved in number naming RAN tasks show deficits in MCI [23‒25]. The Mobile Universal Lexicon Evaluation System (MULES) is a RAN task for picture naming. The Staggered Uneven Number (SUN) is a RAN task for number naming. Both were developed by our group [26‒28]. Previous work shows that the MULES and SUN differentiate patients with MCI from controls in their original paper/pencil format [29]. A recent tablet-based version of the MULES and SUN was developed, called the Mobile Integrated Cognitive Kit (MICK) app [30]. However, the MICK app is not yet investigated for its ability to distinguish MCI from controls [30]. Adapting digital tools to screen for MCI may enable accessibility [31]. Lastly, LCLA measures the ability to distinguish letters at different contrast levels and is known to be reduced in MCI and AD [32‒36]. Contrast sensitivity deficits are linked to damage in the retina and occipital lobe, where amyloid plaques and tau tangles impair light-dark contrast detection [35].

This study aimed to compare picture naming (MULES), number naming (SUN), and LCLA tests to distinguish MCI from controls, hypothesizing that these tests would effectively differentiate the groups. In addition, we aimed to assess, for the first time, whether tablet-based RAN tasks (MICK) could produce results comparable to their traditional paper-and-pencil formats in an MCI cohort.

Participants were recruited from the NYU Barlow Center memory clinic and the NYU Alzheimer’s Disease Research Center. Inclusion criteria were age over 60 and fluency in English. Exclusion criteria were severe psychiatric or neurological disorders, such as major depression, schizophrenia, other dementias, epilepsy, and significant brain injury. The diagnosis of MCI, with a high likelihood of AD as the underlying cause, was determined using the National Institute on Aging-Alzheimer’s Association (NIA-AA) criteria [2]. This included a Global Clinical Dementia Rating^® (CDR) score of 0.5, which is commonly used to identify MCI, along with evidence of abnormal cognitive testing characterized by an amnestic pattern [37, 38]. Confirmation was further supported by neurodegeneration assessed via MRI or FDG PET-MRI. The diagnoses were confirmed by neurologists. Controls had CDR scores of 0 and no cognitive testing abnormalities. Demographic data were collected via questionnaires, and written informed consent was obtained. Participants received a USD 50 incentive. The study was approved by the NYU Grossman School of Medicine Institutional Review Board.

The MULES picture naming test includes 54 color photographs of everyday objects, food, and animals. It was administered on double-sided laminated 8.5 × 11-inch paper with 30 photos on “Side 1” and 24 photos on “Side 2” (Fig. 1). The SUN number naming test includes numerical digits with variable spacing arranged in a horizontal and zigzag pattern. The number of digits ranged from 0 to 12. The SUN was administered on a single-sided laminated 8.5 × 11-inch paper with 154 numbers (Fig. 1). The participants performed two trials for each test. This method of RAN task testing is referred to as the paper/pencil format.

The MICK app is a tablet application version of the MULES picture naming test and the SUN number naming test [30]. It is important to note that not all participants completed the tablet testing. This depended on the availability of research staff. There were no significant demographic differences between participants who completed tablet testing and those who did not. Exact participant numbers are included in the results.

Participants completed two trials in each format (paper/pencil and tablet). Participants named each picture or number aloud, proceeding from left to right across the rows. The fastest time between the two trials was considered the best time and used for analyses. Participants wore their habitual near corrective lenses.

LCLA testing was performed in a dimly lit room and performed standing 2 m away from the letter chart (Fig. 1). LCLA charts (low-contrast Sloan letter charts, Precision Vision, LaSalle, IL, USA) were used. Two LCLA tests were done: at 2.5% and 1.25% contrast. The numbers of letters identified correctly out of 70 were recorded.

Statistical analyses were conducted using STATA 18.0. For demographic data, Levene’s test was conducted to assess homogeneity of variance. Because variances were equal, ANOVA (for comparing >2 demographic groups) or paired t tests (for comparing = 2 demographic groups) were used to compare the differences between MCI and controls. Demographic variables (age, education, sex) were not controlled for in analysis of testing performance between groups, as no significant group differences were found on ANOVA or t tests. Multivariable linear regression models were employed to assess the independent effects of age, education level, and subject group on each outcome, with statistical significance defined as p < 0.05. A full table of these data can be found in the online supplementary content (for all online suppl. material, see https://doi.org/10.1159/000546451).

Wilcoxon rank-sum was used to compare testing performance on the MULES, SUN, and LCLA tasks between groups. A robust regression was conducted to assess agreement between the paper/pencil and tablet versions of the MULES and SUN tasks.

Receiver operating characteristic (ROC) analyses were used to evaluate the capacity of the MULES, SUN, and LCLA tasks to distinguish MCI from controls. The area under the ROC curve (AUC) was calculated, with a standard cut-off of ≥0.50 letters or seconds selected for its balance between sensitivity and specificity. This cutoff aligns with clinical thresholds, ensuring comparability with existing diagnostic standards [39].

The study included 64 participants: 33 controls (mean age 74.6 ± 5.7 years, range 65.0–90.0) and 31 with MCI (mean age 73.2 ± 6.8 years, range 61.0–89.0). No significant age difference was found between the groups (t (62) = 0.88, p = 0.38). In the control group, age correlated with paper/pencil MULES picture naming performance (r = 0.52), with older participants performing slower. A negative correlation was found between education and tablet SUN number naming performance (r = −0.76), where higher education was linked to faster times. Demographic details are in Table 1, with additional information in the online supplementary content.

Overall, 33 controls and 31 MCI patients completed the paper/pencil versions, while 14 controls and 17 MCI patients did the tablet MULES and SUN. For paper/pencil MULES, MCI patients had slower median times (74.96 s, range 49.1 s–211.8 s) compared to controls (53.5 s, range 34.7 s–158.5 s), p < 0.0001. This difference remained significant after controlling for age and education, p < 0.01. On the tablet MULES, MCI patients also had slower times (69.0 s, range 50.5 s–117.2 s) than controls (50.2 s, range 42.4 s–103.0 s), p = 0.01 (Fig. 2). This difference remained significant after controlling for age and education, p = 0.02. Robust regression showed significant agreement between the paper/pencil and tablet MULES (coefficient = 0.80, 95% CI: 0.63, 0.97), p < 0.001. ROC analysis for the paper/pencil MULES showed an AUC of 0.77 (SE = 0.06; 95% CI: 0.65, 0.89), and for the tablet MULES, an AUC of 0.77 (SE = 0.10; 95% CI: 0.58, 0.97). Using a cutoff of ≥50.1 s for the paper/pencil MULES, the sensitivity was 97%, and specificity was 37%. For the tablet MULES, with a cutoff of ≥50.5 s, the sensitivity increased to 100%; specificity was 50% (Fig. 3).

For the paper/pencil SUN, the median score for MCI patients was 59.9 s (range: 42.1 s–74.7 s), compared to 59.5 s (range: 42.1 s–88.0 s) for controls, p = 0.07, showing no significant difference. On the tablet SUN, MCI patients had slightly slower times (median 59.3 s, range: 45.5 s–87.5 s) compared to controls (median 55.7 s, range: 44.5 s–89.1 s), p = 0.36, with no significant group difference (Fig. 4). Robust regression analysis showed strong agreement between the two versions of the SUN task, with a coefficient of 0.97 (95% CI: 0.81, 1.14), p < 0.001. The ROC analysis for the paper/pencil SUN task showed an AUC of 0.63 (SE = 0.07; 95% CI: 0.50, 0.77), and for the tablet SUN, an AUC of 0.60 (SE = 0.10; 95% CI: 0.39, 0.80), indicating modest ability to differentiate MCI patients from normal controls. Using a cutoff of ≥51 s for paper/pencil SUN, the sensitivity was 74% and the specificity was 36%. For the tablet SUN, with a cutoff of ≥55 s, the sensitivity was 65%, specificity was 47% (Fig. 3). All Wilcoxon rank sum results reported are also provided in Table 2.

MCI patients identified a median of 33 letters correct (range: 0–50) on the 2.5% LCLA test, compared to 36 letters (range: 16–53) for controls, p = 0.04*. This difference remained when adjusting for age and education, p < 0.01. In the 1.25% LCLA test, MCI patients identified a median of 0 letters (range: 0–25), while controls identified a median of 14 letters (range: 0–32), p < 0.0001* (Fig. 5). This difference remained when adjusting for age and education, p < 0.01. ROC analysis for 2.5% contrast LCLA showed an AUC of 0.65 (SE = 0.07, 95% CI: 0.52, 0.79). At a cut-off of ≥50 letters, the sensitivity was 97% and specificity was 37. For the 1.25% contrast LCLA, the AUC was 0.72 (SE = 0.06; 95% CI: 0.60–0.84). At a cut-off of ≥50 letters, the sensitivity was 6%, specificity was >99% (Fig. 3).

Overall, the results of this study demonstrate that picture naming (MULES) is a more discriminatory test for MCI versus controls compared to number naming (SUN) and LCLA. Additionally, we show that tablet-based administration of picture naming tasks offers a practical and accessible tool for screening.

Our results align with previous studies that suggest picture naming tasks like the Boston Naming Test and the Multilingual Naming Test are effective screening tools for MCI [17‒20, 29]. The MULES picture naming task uses photographs of real-world objects, unlike the Boston Naming Test and Multilingual Naming Test, which use illustrations. Photographs may engage higher-level visual areas like the fusiform gyrus and temporal lobes, involved in object recognition and memory, whereas illustrations mainly activate basic visual processing regions [40]. This real-world relevance may make the MULES task more sensitive in detecting early cognitive decline.

The ROC analysis of the paper/pencil MULES picture naming task aligns with previous findings, showing similar AUC values (AUC = 0.79 in previous work vs. AUC = 0.77 in this study) [29]. The similarity between the two AUC values suggests that the task is reliable and robust across different cohorts, enhancing its potential as a reliable screening tool.

It is important to consider how vision-based assessments can complement current measures of MCI and AD. Traditional biomarkers of MCI due to AD include post-mortem neuropathological quantification of Aβ plaques, neurofibrillary tangles, and neuritic plaques [41]. In addition, in vivo biomarkers can be quantified using PET, MRI, and CSF analyses [42, 43]. These tests are sensitive but expensive and impractical for routine clinical use.

Alternatively, the Montreal Cognitive Assessment (MoCA) and the Mini-Mental Status Exam (MMSE) are easy to administer and effective in distinguishing MCI patients from controls [44‒46]. In a previous meta-analysis, the discriminatory ability of the MoCA (AUC = 0.85, sensitivity = 80.48%) and MMSE (AUC = 0.74, sensitivity = 66.34%) was evaluated for MCI versus controls using the task’s respective published/recommended cut-offs for MCI detection [46]. Our current findings show that the MULES tasks (AUC = 0.77, sensitivity = 97–100%) have a similar discriminatory power to the MoCA and MMSE in a significantly smaller cohort and offer significantly higher sensitivity at a 50-letter cut-off score. While the MULES may result in more false positives, prioritizing sensitivity ensures that fewer cases of MCI are overlooked.

This study has several strengths. To our knowledge, this is the largest cohort study to directly compare picture naming, number naming, and LCLA in MCI. Additionally, it is the first to explore the tablet-based administration of the MULES and SUN, demonstrating their potential as efficient, accessible screening tools. Furthermore, MCI patients and controls in this study had similar ages and education levels, helping to isolate testing differences due to cognitive differences from external factors.

This study had several limitations. The small sample size limits generalizability and the ability to detect group differences, particularly for tablet RAN tasks. Performance bias might also arise from prior exposure to cognitive assessments, influencing test results. In contrast, participants with less technological experience may have struggled with the tablet-based testing. Additionally, self-selection bias could have occurred, as participants who were more motivated or health-conscious may have been more likely to participate. Furthermore, two significant correlations were observed that could influence the results: older age was linked to slower paper/pencil MULES times, and higher education was associated with faster tablet SUN times. While differences in demographics between groups might affect the results, these correlations are likely due to chance because they only appeared in one format (paper/pencil or tablet) and not across both. To improve the accuracy of future findings, research with a larger sample should account for age and education differences when comparing performance between test formats. This will help clarify the diagnostic potential of these tasks.

Future research should examine how cognitive reserve, including pre-morbid IQ and education, impacts performance on picture naming tasks like the MULES. Additionally, directly comparing performance on the MMSE and MoCA with picture naming can better assess their comparative discriminatory capacity for MCI. Future studies could also compare fMRI or PET biomarkers with MULES picture naming performance over time, to clarify when decline on vision-based assessments correlates with measurable brain changes. Finally, follow-up studies should compare MULES picture naming with the Boston Naming Test or Multilingual Naming Test to explore differences in distinguishing MCI.

In conclusion, our study highlights the potential of picture naming tasks as accessible and early indicators of MCI due to AD in both original paper/pencil and tablet formats. By offering a practical and cost-effective alternative to traditional biomarkers, these tests could serve as valuable tools in routine clinical settings for early detection.

The authors confirm that all research procedures followed ethical guidelines, including obtaining informed consent from all participants, ensuring data confidentiality, and adhering to relevant institutional review board regulations. This study protocol was reviewed and approved by the NYU Grossman School of Medicine Institutional Review Board, Approval No.: i19-01219. The study was conducted in accordance with local and national guidelines and adhered to the principles outlined in the Declaration of Helsinki. Written informed consent was obtained from all participants prior to their inclusion in the study. In the case of minors, consent was obtained from a parent/legal guardian. The study protocol and consent procedures were approved by the as part of the ethical review process.

Dr. Arjun V. Masurkar is a council member of the Alzheimer’s Association International Research Grant Program and a steering committee member of the Alzheimer’s Disease Cooperative Study and serves on the Editorial Board of Alzheimer’s and Dementia: Translational Research and Clinical Interventions. The authors declare no conflicts of interest. The authors affirmed that these affiliations did not influence the content or integrity of this manuscript.

This study was supported by the NYU Grossman School of Medicine and funded by NIH grants P30AG066512 and OT2OD038130.

L.S. conceived the study, performed data analysis, and drafted/revised the manuscript. S.H. assisted with data analysis and participant enrollment. R.K. contributed to the statistical analysis and manuscript editing. A.N. contributed to literature review. A.V.M. provided oversight and medical expertise and contributed to the manuscript revisions. S.G. also provided oversight and medical expertise and contributed to the manuscript revisions. L.B. provided medical expertise, contributed to data analysis, manuscript revision, and study design. L.B. is also the creator of the MULES and the SUN used in this study.

The data that support the findings of this study are not publicly available due to participant privacy but are available from the corresponding author upon reasonable request.