A Complementary Framework for Early Cognitive Screening: Targeting Medial Temporal and Retrosplenial Circuits Underserved by Speed-Based Digital Batteries
A theoretical framework for complementary evaluation using a rotational spatial memory task targeting the medial temporal lobe and retrosplenial (posterior cingulate) cortex
Yoshihiko Suzuki
MyMedipro Inc. / HDC Atlas Clinic
Abstract Background and objective. Processing-speed–oriented digital cognitive assessments, represented by the Cogstate Brief Battery (CBB; marketed in Japan as the non-medical device “nō-KNOW”), quantify psychomotor function, attention, working memory, and visual learning within a short time and have been reported to discriminate healthy individuals from those with cognitive decline with high sensitivity and specificity. Their measurement center of gravity, however, lies in the frontoparietal processing-speed system, and their sensitivity to the medial temporal lobe and retrosplenial cortex (posterior cingulate) circuits that Alzheimer’s disease (AD) pathology damages earliest may be structurally limited. This paper derives, from existing neuroscientific knowledge, the hypothesis that a rotational spatial paired-associate memory task (working title “Guru-Guru Memory”) can complement this spatial blind spot at the level of neural substrate. Methods and structure. We organize the intracerebral progression of AD pathology (Braak staging; early degeneration of the entorhinal cortex, hippocampus, and retrosplenial cortex) and map the four CBB tasks onto a cognitive-domain–to–neural-substrate correspondence table. We then decompose the present task into four cognitive components and juxtapose their putative neural substrates, thereby showing the complementary structure of the two assessments. Finally, we present the working hypothesis of incremental validity—that a composite index integrating both may have discriminative power for the earliest cognitive decline exceeding that of either alone—and propose a falsifiable study design to test it. Conclusion. The present task primarily targets the medial temporal–retrosplenial circuit that speed-based batteries underserve and may supply information orthogonal to existing tools. Their integration may, in theory, raise the discriminative power for the earliest cognitive decline. At present, however, all of these claims are unverified hypotheses that require empirical demonstration of construct and incremental validity through normative-data development and case-control studies. |
Keywords: early Alzheimer’s disease; medial temporal lobe; retrosplenial cortex; pattern separation; spatial navigation; digital biomarker; incremental validity
1. Introduction
As disease-modifying therapies for dementia become realistic options, there is a growing need to capture, by simple and repeatable means, the mild cognitive impairment (MCI) or preclinical stage in which the effect of intervention is most expected. In this context, short digital assessments—computerized versions of paper-and-pencil tests—have spread rapidly. A representative example is Cogstate’s Cogstate Brief Battery (CBB), marketed in Japan by Eisai as the non-medical device “nō-KNOW.” The CBB has been reported to detect cognitively impaired groups with a sensitivity of approximately 80% and specificity of approximately 85% on a memory index computed from visual learning and working memory, and the same algorithm is used abroad as a medical device (Cognigram) to support the diagnosis of MCI and dementia.
Every cognitive assessment selectively reflects the brain networks that correspond to the processing its task design demands. The four CBB tasks—psychomotor function, attention, working memory, and visual learning—are all processing-speed–weighted in composition, with reaction time as the principal dependent variable. This is an excellent design choice that guarantees high reproducibility and low noise; at the same time, for the kinds of early impairment that do not manifest strongly in reaction time—particularly impairment of associative memory in the medial temporal lobe and of spatial reorientation in the retrosplenial cortex—its sensitivity can, in principle, be blunted.
The aim of this paper is to derive theoretically, from existing neuroanatomical and neuropsychological knowledge, the hypothesis that a rotational spatial paired-associate memory task, “Guru-Guru Memory,” can complement this structural blind spot. In addition, we present the working hypothesis of incremental validity—that a composite index integrating both assessments may have early-discrimination power exceeding that of either alone—and make explicit the procedure for its falsification. We state at the outset that this paper is a theoretical, hypothesis-generating discussion and is not intended to present empirical data.
2. Early Progression of Alzheimer’s Disease Pathology and the Brain Map
The neurofibrillary changes of AD (tau pathology) do not spread randomly but progress in a relatively stereotyped order, as shown by Braak and colleagues. The earliest stage (Braak stages I–II) affects the entorhinal cortex and transentorhinal region, followed by the medial temporal lobe including the hippocampus and parahippocampal gyrus, then the temporoparietal association cortex, and finally the entire neocortex except the primary sensorimotor areas. In other words, the medial temporal lobe system that supports memory and spatial cognition is, pathologically, the region damaged earliest and most sensitively.
Functional imaging findings agree with this. On FDG-PET, hypometabolism in the posterior cingulate and retrosplenial cortex (the precuneus/retrosplenial region) is one of the most reproducible findings, consistently observed from relatively early in AD. The retrosplenial cortex is a hub that interconverts egocentric coordinates (apparent positional relationships) and allocentric coordinates (position within the world) and reorients the sense of direction; functional decline in this region manifests clinically as the spatial disorientation characteristic of early AD—“getting lost in familiar places.”
Furthermore, the cognitive-map system formed by grid cells in the entorhinal cortex and place cells in the hippocampus underlies the metric representation of space and its sequential updating during movement (path integration / spatial updating). Both animal models and human functional imaging suggest that the function of this system can deteriorate from the early stages of AD pathology. In short, the core circuit of the earliest AD is the “medial temporal lobe (entorhinal cortex, hippocampus) plus retrosplenial cortex,” and its impairment first reveals itself not as slowing of reaction speed but as impairment of associative memory, pattern separation, and spatial updating.
3. Measurement Domains and Structural Limitations of the Cogstate Brief Battery (nō-KNOW)
Table 1 organizes the four CBB tasks, the cognitive function each principally demands, and the putative neural substrate. Every task uses, as its principal index, reaction time premised on accurate responding, or the learning speed of visual association over a short interval.
Table 1. Task composition and (putative) neural substrates of the Cogstate Brief Battery
| Task | Principal cognitive function | Putative neural substrate | Principal index |
|---|---|---|---|
| Detection | Psychomotor speed | Frontal–subcortical; white-matter integrity | Reaction time |
| Identification | Selective attention / vigilance | Frontoparietal attention network | Reaction time |
| One Card Learning | Visual recognition learning | Medial temporal lobe (in part); visual association cortex | Accuracy |
| One Back | Working memory | Dorsolateral prefrontal cortex; parietal lobe | Reaction time / accuracy |
Note. Task names follow the general composition of the CBB. Neural substrates are estimates from representative findings; each task requires the coordination of multiple regions.
What Table 1 shows is that the CBB places its measurement center of gravity on the frontoparietal processing-speed, attention, and working-memory systems. One Card Learning does partially load the medial temporal lobe system, but its format is “recognition of the identical stimulus” and does not extend to the pattern separation discussed below (discrimination among similar stimuli) or to the updating of spatial representation during movement. Therefore, the earliest-AD circuit described in Section 2—the pattern-separation function of the entorhinal cortex and the spatial-reorientation function of the retrosplenial cortex—is not sufficiently contained within the task structure of the CBB. This is not a defect but an inevitable consequence of a design that prioritizes reproducibility; nonetheless, in light of the goal of early detection, it leaves a structural blind spot.
4. Cognitive–Neural Decomposition of the Rotational Spatial Paired-Associate Memory Task (Guru-Guru Memory)
The present task is not a single memory task but a composite of multiple cognitive components with distinct neural substrates. To interpret a decline in score, one must first decompose the task into components. The present task can be decomposed into the following four components.
Component 1: Object–location binding
This is the part in which the participant encodes and recalls, on a 3×3 grid, “which number was in which cell” and “where were the two cells that contained a matching pair of the same number.” This is associative memory that binds item information (the number) with spatial context (the location), and it depends strongly on the hippocampus (particularly CA3 and CA1). The hippocampus is the very device that integrates item and context; when this function declines, the characteristic error arises of remembering the number yet mistaking its location.
Component 2: Delayed associative recall across interference
Between encoding the grid and recalling the location, a delay and the entire concentration (memory) game intervene as an interference task. The recall task thus becomes “delayed associative recall resistant to interference.” This depends on the hippocampus and medial temporal lobe (including the entorhinal cortex), and a pattern of vulnerability to interference with rapid forgetting is a core finding of amnestic MCI and early AD. Among the components of this task, the signal most likely to be sensitive to the earliest AD pathology is thought to reside here.
Component 3: Spatial updating under rotation (the core of “Guru-Guru”)
Each time an incorrect response is made, the board rotates in a stepwise fashion, and the participant must mentally re-track the new position of each card after rotation. This processing mobilizes multiple posterior regions: first, the grid-cell system of the entorhinal cortex (spatial metric and updating); second, the retrosplenial cortex (conversion between egocentric and allocentric coordinates and reorientation of direction); and third, the posterior parietal cortex (intraparietal sulcus and superior parietal lobule; mental rotation itself). As described in Section 2, the entorhinal and retrosplenial cortices are both early targets of AD pathology, and this component loads the same neural circuit as clinical spatial disorientation. This is the core that most clearly differentiates the present task from existing tools.
Component 4: Executive function, attention, and processing speed
This includes strategies for solving the concentration game efficiently, inhibitory control (not flipping known non-matching cards), and flipping speed. It corresponds to the frontoparietal control system including the dorsolateral prefrontal cortex and anterior cingulate cortex, and to processing speed (including white-matter integrity and vascular contributions). This component partially overlaps with the CBB, but in the present task it is scored independently as “wasted flips” and “time required,” and is separable from the memory and spatial indices.
What is important is that, through Components 1–3, the present task biases its measurement center of gravity toward the medial temporal–retrosplenial system. This coincides precisely with the region the CBB underserves. Table 2 juxtaposes the measurement centers of gravity of the two assessments.
Table 2. Measurement centers of gravity of the two assessments and their correspondence to early-AD target circuits
| Brain region / circuit | Early-AD target status | CBB (nō-KNOW) | Guru-Guru Memory |
|---|---|---|---|
| Entorhinal cortex (grid cells) | Earliest (Braak I–II) | △ weak | ◎ strong |
| Hippocampus (association / pattern separation) | Early | △–○ partial | ◎ strong |
| Retrosplenial cortex / posterior cingulate | Early (FDG-PET) | △ weak | ○–◎ moderate–strong |
| Dorsolateral prefrontal cortex / attention system | Middle–late | ◎ strong | ○ moderate |
| Processing speed / white matter | Non-specific | ◎ strong | ○ moderate |
Note. ◎ = principal target; ○ = loaded appropriately; △ = limited. The correspondences are theoretical estimates and are not based on actual measurement.
5. Spatial Complementarity — How to Cover the Blind Spot of the CBB
The central argument derived from Table 2 is clear. The measurement centers of gravity of the CBB and Guru-Guru Memory are arranged almost complementarily along the axis of early-AD target circuits. The CBB is strong in the frontoparietal processing-speed, attention, and working-memory systems and weak in the medial temporal–retrosplenial system; Guru-Guru Memory is the reverse. The crux of this paper is the expectation that the two are not in a redundant relationship measuring the same latent trait from different angles, but in a partially orthogonal relationship (containing mutually independent information).
This complementarity has meaning beyond mere domain coverage. What deteriorates first in the earliest AD is predicted to be not the processing speed that is the CBB’s main arena, but the medial temporal–retrosplenial function that is Guru-Guru Memory’s main arena. Therefore, in the very earliest stage of disease, even while the CBB still shows values within the normal range, the spatial and associative indices of Guru-Guru Memory may deviate ahead of it—such a temporal dissociation is theoretically possible. If this dissociation actually exists, it would be direct grounds for the clinical value of using the two together.
6. Theoretical Inference of Incremental Validity from Integration
On the basis of the above, we formulate the central working hypothesis of this paper as follows.
Working hypothesis (incremental validity) A composite model that adds the medial temporal–retrosplenial indices derived from Guru-Guru Memory (Components 1–3) to the processing-speed and working-memory indices derived from the CBB will, compared with the CBB-alone model, statistically significantly improve the discriminative power (ROC-AUC) for MCI / preclinical AD. |
The theoretical grounds for this hypothesis can be organized into three points. First, because the two indices reflect different neural substrates, they are expected to have partial statistical orthogonality, and the variance unexplained by one may be explained by the other (room for an incremental contribution). Second, because the earliest AD pathology that is the target of discrimination affects the system Guru-Guru Memory mainly measures before the system the CBB mainly measures, the additional index may work advantageously from the standpoint of sensitivity. Third, combining multiple quasi-independent indices averages out the measurement error of a single index and increases the stability of the discrimination boundary (from the standpoint of reliability).
This hypothesis carries clear falsification conditions. If the two indices are highly correlated (redundant) and the Guru-Guru Memory index is largely subsumed within the explained variance of the CBB index, the incremental AUC will converge to zero and the hypothesis will be rejected. Moreover, if the measurement noise introduced by the rotation operation (confounds of mental-rotation ability, visuospatial ability, and operational proficiency) exceeds the signal, the additional index may instead worsen discrimination. It is therefore important that this hypothesis is formulated not as mere expectation but in a form falsifiable by a single quantity, the incremental AUC.
7. Proposed Validation Design
We present a minimal study design for testing the above hypothesis. The objective is to determine, in a falsifiable form, whether the Guru-Guru Memory index has incremental validity over the CBB.
Participants will be a healthy group and an MCI group, stratified by specialist clinical diagnosis (NIA-AA criteria), with plasma biomarkers (e.g., p-tau217) or imaging (hippocampal volume; retrosplenial metabolism by FDG-PET) used concurrently as reference standards to the extent feasible. All participants will undergo the CBB and Guru-Guru Memory on the same occasion; the latter will be recorded not as a total score but decomposed into the component subscores of Section 4 (object–location binding, delayed recall, post-rotation updating, and execution/speed).
Analysis will use hierarchical logistic regression. In the first step, the discriminative power (AUC) of a model entering only the CBB indices will be estimated; in the second step, the Guru-Guru Memory subscores will be added, and the increment in AUC and its significance by likelihood-ratio test will be evaluated. In addition, as confirmation of construct validity, the correlations between the post-rotation updating subscore and retrosplenial metabolism, and between the delayed-recall subscore and hippocampal volume, will be examined. In the longitudinal portion, both assessments will be administered repeatedly and the power to detect within-individual change (the Reliable Change Index, RCI) will be compared. Because the high-frequency repetition enabled by the game format may provide an index sensitive to early change—namely the slope from baseline—the discriminative power of the longitudinal slope will be evaluated in addition to cross-sectional comparison of absolute values.
The sample size must be designed in advance based on the power to detect an assumed moderate incremental effect; because securing it at a single site is generally difficult, the study presupposes collaborative research with institutions that have memory-disorder clinics (with ethics-committee approval and pre-registration).
8. Limitations
All claims of this paper remain at the stage of theoretical inference. First, the correspondences between brain regions and subscores presented here are estimates from representative findings and are not based on actual measurement using the present task; because each component requires the coordination of multiple regions, attribution to a single region is only an approximation. Second, the working hypothesis of incremental validity is unverified, and the possibilities that the two indices are redundant and that rotation-derived noise degrades the signal cannot be excluded. Third, the present task currently lacks normative data, actual measurement of reproducibility, and procedures for controlling confounds such as years of education, device differences, visual acuity, and motor speed. Fourth, individual differences in motivation and proficiency associated with the game format complicate its interpretation as a clinical index. Because of these limitations, the present task should at this stage be positioned as recreational cognitive training and is not at a stage where diagnostic or screening performance can be claimed.
9. Conclusion
Processing-speed digital cognitive assessments, while possessing high reproducibility and discriminative performance, may—because of their measurement center of gravity—leave a structural blind spot with respect to the medial temporal–retrosplenial circuit that the earliest AD pathology damages. The rotational spatial paired-associate memory task targets the very center of this blind spot through its components of object–location binding, delayed recall under interference, and post-rotation spatial updating. The theoretical conclusion of this paper is that the two are arranged complementarily along the axis of early-AD target circuits and that their integration may raise the discriminative power for the earliest cognitive decline. However, this conclusion is a hypothesis falsifiable by a single quantity, the incremental AUC, and can be elevated to a clinical claim only after empirical demonstration through normative-data development and collaborative case-control studies. That empirical demonstration is the sole path that bridges the present task from recreation to clinical application.
Principal References
1) Braak H, Braak E. Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol. 1991;82(4):239–259.
2) Maruff P, Lim YY, Darby D, et al.; AIBL Research Group. Clinical utility of the Cogstate Brief Battery in identifying cognitive impairment in mild cognitive impairment and Alzheimer’s disease. BMC Pharmacol Toxicol. 2013;14:30.
3) Minoshima S, Giordani B, Berent S, et al. Metabolic reduction in the posterior cingulate cortex in very early Alzheimer’s disease. Ann Neurol. 1997;42(1):85–94.
4) Yassa MA, Stark CEL. Pattern separation in the hippocampus. Trends Neurosci. 2011;34(10):515–525.
5) Coughlan G, Laczó J, Hort J, Minihane AM, Hornberger M. Spatial navigation deficits — overlooked cognitive marker for preclinical Alzheimer disease? Nat Rev Neurol. 2018;14(8):496–506.
6) Vann SD, Aggleton JP, Maguire EA. What does the retrosplenial cortex do? Nat Rev Neurosci. 2009;10(11):792–802.
7) Kunz L, Schröder TN, Lee H, et al. Reduced grid-cell–like representations in adults at genetic risk for Alzheimer’s disease. Science. 2015;350(6259):430–433.
8) Jack CR Jr, Bennett DA, Blennow K, et al. NIA-AA Research Framework: Toward a biological definition of Alzheimer’s disease. Alzheimers Dement. 2018;14(4):535–562.
Note: This paper is a theoretical, hypothesis-generating discussion; the performance and validity described are all unverified hypotheses. The present task is not a test for medical purposes and is not intended for diagnostic or screening use.