The spatial learning task of Lhermitte and Signoret: Marker of bilateral hippocampal dysfunction in temporal lobe epilepsy

Abstract

Objective

Bilateral hippocampal compromise in temporal lobe epilepsy (TLE) substantially increases the risk of significant adverse memory outcomes following unilateral anterior temporal lobectomy and may contraindicate surgery. The Spatial Learning Task (SLT) of Lhermitte and Signoret (1972) is a simple object–location arbitrary associative learning task proposed as a marker of bitemporal dysfunction. Here, we test whether the SLT might serve as a useful marker in TLE by examining, in individuals with TLE and hippocampal sclerosis (HS), whether performance on the SLT depends on the laterality of the HS (left, right, bilateral). We hypothesize that performance will be worst in the setting of bilateral HS.

Methods

Retrospective analysis of 63 surgically naïve TLE‐HS patients (16 bilateral HS, 28 left HS, 19 right HS) who underwent neuropsychological evaluation at a tertiary epilepsy center.

Results

Bilateral HS cases showed significantly poorer SLT learning and learning efficiency compared to right HS cases (p ≤ .005, BF₁₀ ≥ 25). Left HS cases performed comparably to bilateral HS cases on early trials but typically achieved ceiling performance with repetition, reflecting inefficiency rather than capacity loss. An SLT learning score <13 yielded high specificity (96% [95% CI: 0.85–0.99]) but modest sensitivity (44% [95% CI 0.20–0.70]) for identifying bilateral HS cases.

Significance

In the setting of TLE‐HS, poor performance on the SLT increases the probability of bilateral hippocampal pathology, supporting its utility as a rule‐in indicator. Although intact scores do not reliably exclude bilateral pathology, a low score on this task should prompt further investigations as to possible bitemporal compromise, and suggest risk of adverse memory outcomes following surgery. Prospective, post‐operative studies will be required to quantify this risk.

Plain Language Statement

The results of the study show that the ability to learn arbitrary object–location pairings is compromised in the setting of temporal lobe epilepsy with bilateral hippocampal sclerosis (HS). In contrast, patients with unilateral HS generally retained sufficient capacity to learn the task, though those with left HS were slower to acquire the pairings across repeated trials. A low score on this task was a strong indicator of bilateral hippocampal pathology, whereas an intact score did not reliably exclude it. Our results suggest that performance on this simple learning task provides a useful marker of bilateral hippocampal compromise in epilepsy and may be particularly valuable in patients for whom standard memory tests are less suitable. We provide a cutoff score to facilitate clinical interpretation.

Key points.

• Patients with temporal lobe epilepsy with bilateral hippocampal sclerosis (HS) showed poorer performance on the Spatial Learning Task compared to unilateral HS.

• A learning score below 13 across the first three trials was highly specific for bilateral hippocampal involvement, though sensitivity was modest.

• Patients with unilateral HS demonstrated slower initial learning but typically reached ceiling with repetition, reflecting reduced learning efficiency rather than loss of capacity.

1. INTRODUCTION

Following Scoville and Milner's seminal publication, the profound amnesia seen after bilateral anterior temporal lobectomy (ATL) was occasionally also observed in patients who had undergone unilateral ATL. ¹ , ² , ³ In such cases, evidence of occult pathology involving the contralateral (unoperated) hippocampus emerged or was suspected. To avert these catastrophic memory outcomes, characterization of hippocampal integrity became an imperative in the presurgical work‐up of patients with temporal lobe epilepsy (TLE).

Currently, a range of investigations are used to probe for signs of bilateral hippocampal compromise. These include structural and functional findings such as bilateral hippocampal atrophy on MRI, bitemporal hypometabolism on PET, or EEG evidence of seizures arising independently in both temporal lobes. ⁴ , ⁵ , ⁶ Importantly, however, surgical risk hinges on the functional integrity of memory networks, which cannot be inferred from these measures alone; direct neuropsychological assessment remains crucial. ⁷ In this regard, the material‐specific model of memory—which posits verbal and non‐verbal learning and memory to be sensitive to the integrity of the left and right hippocampi respectively—guides memory evaluation in most surgical epilepsy programs, with decrements across both modalities interpreted as evidence of bilateral hippocampal dysfunction. However, material‐specific dissociations are imperfect and may fail to resolve laterality. ⁸ This motivates the need for a simple, non‐lateralising memory measure that is sensitive to bilateral hippocampal dysfunction.

One candidate is the Spatial Learning Task (SLT) of Lhermitte and Signoret, ⁹ an elementary associative memory task that involves the learning and recall of nine arbitrary object–location pairings. Of clinical interest is that, in principle, this task can be completed using either verbal and/or non‐verbal strategies. ¹⁰ For example, participants may verbally label object–location pairs using grid‐based heuristics (e.g., “CHAIR—MIDDLE”; so‐called categorical encoding), or alternatively, encode positional coordinates nonverbally (coordinate encoding). Consistent with this idea, similar paradigms have failed to yield laterality effects in left versus right TLE, ¹¹ suggesting that the SLT may be performed normally even in the setting of unilateral hippocampal dysfunction. Performance also appears relatively unaffected by demographic or cognitive factors such as age, education, or IQ, ¹⁰ raising its potential value in diverse or difficult‐to‐assess populations. While early reports in non‐epilepsy populations linked poor SLT performance to bilateral hippocampal dysfunction, ⁹ , ¹² its value as a clinical marker of bitemporal involvement in TLE has not been examined.

To assess the SLT's potential as a marker of bitemporal dysfunction, we used MRI‐detected HS as an anatomical index of hippocampal compromise, comparing SLT performance across patients with bilateral, left, or right HS. We hypothesized that patients with bilateral HS would perform more poorly on the SLT than those with unilateral HS. In addition, we examined the task's clinical utility by quantifying its diagnostic accuracy (e.g., sensitivity, specificity) for detecting bilateral hippocampal involvement.

2. METHODS

2.1. Participants

The study included 63 surgically naïve patients with TLE and HS (TLE‐HS: 16 bilateral HS, 28 left HS, 19 right HS). HS laterality was assigned on brain MRI following neuroradiology review, based on standardized criteria for visual assessment. ¹³ No patients with dual pathology (i.e., HS co‐occurring with another potentially epileptogenic lesion) were included.

The inclusion criteria were patients aged 18 years or older, with available scores on the SLT. ⁹ Cognitive data were acquired as part of a clinically directed neuropsychological evaluation in the Bladin‐Berkovic Comprehensive Epilepsy Program (CEP), Austin Health, Melbourne, between 1990 and 2024. Epilepsy categorization during the CEP admission typically included derivation of a detailed clinical history, inpatient video‐EEG monitoring, structural brain MRI, interictal [¹⁸F] fluorodeoxyglucose PET, together with ictal and interictal single‐photon emission computed tomographic (SPECT). Lateralization of the seizure focus was determined by CEP consensus review integrating these multimodal data. This study was approved by the Austin Health Human Research Ethics Committee (HREC: 88149). The requirement for individual informed consent was waived given the retrospective design and use of de‐identified data.

2.2. The spatial learning task

The SLT ⁹ assesses the ability to learn arbitrary associations between objects and their spatial locations. The task comprises nine stimulus cards depicting line drawings of everyday objects, each occupying a fixed position within a 3 × 3 grid. During learning, the examiner presents the nine object‐location pairings in random order, instructing the participant to remember each location. Each card is then presented individually, and the participant indicates its location; feedback is provided after each response. On each learning trial, all nine picture‐location associations are re‐presented. Under standard administration, testing proceeds until the participant places all cards correctly on two consecutive trials or completes 10 learning trials, whichever occurs first. Following an approximately 30‐min delay, participants complete a delayed recall trial in which each object card is presented individually, and the participant indicates its corresponding location.

In this dataset, testing was often truncated due to clinical constraints as extended testing beyond the initial trials was not deemed clinically warranted. All patients completed at least two trials, with the majority (n = 48/63; 77.8%) completing at least three trials, and a smaller subset (n = 30; 47.6%) completing four or more trials.

To operationalize SLT performance, one primary index (“SLT Learning”) and four secondary indices were derived:

1. “SLT Learning”: Total score across the first three trials (range: 0–27). 14 of 63 participants attained a perfect score (9/9) on Trial 2 and were not administered a third learning trial; in these cases, Trial 2 performance was carried forward to impute Trial 3 (i.e., these cases were assigned a score of 9/9 for Trial 3), enabling consistent calculation of this summary score. This carry‐forward assumption was supported empirically: among participants who achieved a ceiling score (9/9) on Trial 2 but were nonetheless administered a genuine Trial 3, all again scored 9/9, with no instance of regression from an established Trial 2 ceiling.

2. “SLT Delay”: Number of correctly recalled items following a 30‐min delay (range: 0–9), reflecting delayed recall.

3. “SLT Peak”: Maximum score achieved across all completed learning trials (range: 0–9), reflecting peak performance regardless of when it was achieved.

To account for variability in the number of trials administered, two complementary learning trajectory indices were computed:

• 4“Relative Gain”: Improvement from the first to the final administered learning trial, scaled to the maximum possible improvement, ranging from 0 (no improvement) to 1 (maximum possible improvement given the starting score).

$$\text{Relative Gain}=\frac{\text{Final Score}-\text{Trial}1\text{Score}}{9-\text{Trial}1\text{Score}}$$

• 5 “Learning Efficiency (AULC)”: Overall efficiency of learning across all trials, expressed as the proportion of the maximum possible learning curve for the same number of trials, ranging from 0 (no learning) to 1 (perfect performance on every trial).

$$\text{Learning Efficiency}=\frac{\text{Observed AULC}}{\text{Maximum Possible AULC}}$$

where

• Observed AULC = (S ₁ + S ₂)/2 + (S ₂ + S ₃)/2 + … + (S _n−1 + S _n )/2

• S _i  = score on trial i

• n = number of trials administered

• Maximum Possible AULC = (n − 1) × 9

2.3. Statistical analysis

Analyses were conducted in R (version 4.3.2), with α = 0.05 to determine statistical significance. Both frequentist and Bayesian methods were employed to examine the robustness of findings. Bayesian analyses used BayesFactor ¹⁴ (default priors unless specified). Bayes factors (BF₁₀) are interpreted as the ratio of marginal likelihoods of the data under the alternative versus the null: BF₁₀ ≥ 3 = substantial evidence for the alternative; BF₁₀ ≤ 1/3 = substantial for the null; BF₁₀ between 1/3 and 3 = inconclusive. ¹⁵

Variables were screened for missing data and parametric assumptions. Normality of residuals was examined with Q–Q plots and homoscedasticity with Levene's test. Group differences in demographics were tested with Kruskal–Wallis (continuous) and χ ² or Fisher's exact (categorical). As indices derived from the SLT were positively skewed, the non‐parametric Kruskal–Wallis test was used to compare MRI‐defined HS groups (left, right, bilateral). Post‐hoc pairwise comparisons were conducted using Wilcoxon's tests with Holm‐adjusted corrections to adjust for multiplicity.

Classification performance (bilateral HS versus unilateral HS) was examined using receiver operating characteristic (ROC) curves using the pROC ¹⁶ package in R. The area under the curve (AUC) statistic was calculated with 95% CIs via 2000 stratified bootstrap samples. The optimal cutoffs for classifying bilateral HS were determined by maximizing Youden's J statistic. Sensitivity, specificity, negative predictive values (NPV), positive predictive values (PPV), positive likelihood ratios (LR+), and negative likelihood ratios (LR−) were computed.

Individual learning trajectories were categorized using “Relative Gain”, which quantifies relative improvement from the first to the final administered trial, scaling change according to the maximum score attainable. Alternative indices (e.g., “SLT Learning”, “Learning Efficiency”) were considered less appropriate because they conflate initial performance with rate of change. Three learning profiles were defined as follows:

1. Ceiling: Trial 1 score of 8 or 9, indicating little or no capacity for further improvement.

2. Improvers: “Relative Gain” ≥0.40, indicating substantial improvement across trials.

3. Limited Improvers: “Relative Gain” <0.40, indicating limited improvement across trials.

The 0.40 cutoff was chosen as a pragmatic marker of meaningful improvement, informed by clinical judgment and inspection of observed trajectories. Distributions of learning profiles were compared using χ ² tests of independence; standardized residuals identified cells contributing most.

3. RESULTS

3.1. Sample characteristics

Table 1 presents the clinical characteristics of participant groups.

TABLE 1.

Sample characteristics by group.

Variable	Groups			Comparison
	Bilateral HS (N = 16)	Left HS (N = 28)	Right HS (N = 19)	p h	BF10
Sex				>0.9	0.2
Female	7 (44%)	13 (46%)	9 (47%)
Male	9 (56%)	15 (54%)	10 (53%)
Age, years	38.5 (29.0, 46.3)	39.0 (28.0, 50.5)	44.0 (31.0, 51.5)	0.6	0.2
Handedness a				0.7	0.006
Ambidextrous	0 (0%)	2 (7.1%)	0 (0%)
Left	2 (14%)	2 (7.1%)	1 (5.3%)
Right	11 (79%)	23 (82%)	18 (95%)
Age at seizure onset, years a	20.0 (4.0, 25.0)	12.0 (5.0, 18.5)	8.0 (2.0, 24.0)	0.7	0.3
Epilepsy duration, years a	16.0 (8.5, 24.0)	27.0 (17.0, 37.0)	27.0 (24.0, 34.0)	0.06	0.7
Lateralization, seizure focus b				—	—
Bilateral	6 (38%)	0 (0%)	0 (0%)
Left	5 (31%)	27 (100%)	0 (0%)
Right	5 (31%)	0 (0%)	19 (100%)
PET positive c				—	—
PET positive/Concordant	9 (69.2%)	23 (88.5%)	14 (87.5%)
PET negative/Discordant	4 d (30.8%)	3 e (11.5%)	2 f (12.5%)
IQ g	87.0 (73.3, 97.0)	90.0 (81.0, 93.5)	92.0 (84.0, 101.0)	0.6	0.2

Note: Counts (frequencies) presented for discrete variables; medians (interquartile range) presented for continuous variables. PET Concordant/Discordant refers to concordance in relation to laterality of HS.

Abbreviations: BF, Bayes factor; HS, hippocampal sclerosis; IQ, intelligence quotient; TLE, temporal lobe epilepsy.

^a n = 4 absent data.

^b n = 1 unknown lateralization.

^c n = 8 absent data.

^d n = 4 discordant (left temporal hypometabolism).

^e n = 2 discordant (bitemporal hypometabolism), 1 negative.

^f n = 2 discordant (bitemporal hypometabolism).

^g n = 19 absent data (bilateral: 2, left: 9; right: 8). The rate of missing IQ data did not differ significantly across groups (p = 0.17, BF₁₀ = 1.25).

^h Kruskal–Wallis test for age in years, age at seizure onset, epilepsy duration, and IQ, Pearson's chi‐square test for sex, Fisher's exact test for handedness, lateralisation of seizure foci, and PET positive counts.

The groups were broadly comparable on demographic and clinical features, including sex, age at neuropsychological evaluation, handedness, age of seizure onset, epilepsy duration, and estimated IQ. IQ data were unavailable for 19 patients (bilateral: 2/16, 13%; left: 9/28, 32%; right: 8/19, 42%); missingness rates did not differ significantly across groups (p = 0.17, BF₁₀ = 1.25), and available IQ estimates did not differ between groups (p = 0.6, BF₁₀ = 0.2). Notably, bilateral HS showed a trend toward older age of seizure onset and shorter epilepsy duration compared to unilateral cases, but these differences did not reach statistical significance.

Multimodal concordance with MRI‐defined HS laterality was examined across available investigations. All unilateral HS patients showed seizure focus concordant with MRI‐defined HS laterality after CEP consensus review (with laterality remaining unknown in one case of left HS). PET was concordant with MRI in most patients (46/55 with available data); discordant cases included bilateral HS with unilateral (left‐sided) hypometabolism (n = 4), unilateral HS with bitemporal hypometabolism (left HS: n = 2; right HS: n = 2), and one left HS case with no PET abnormality. These findings confirm the expected concordance of seizure focus with MRI‐defined laterality in unilateral HS, while also underscoring the greater heterogeneity of ancillary findings in bilateral HS. A sensitivity analysis restricting the unilateral group to cases with unambiguous lateralization across all available modalities is reported in Supplementary Material A.

3.2. Bilateral HS is associated with lower scores on the SLT

Visual inspection of Figure 1 shows that bilateral HS cases generally performed poorest across all SLT indices, right HS cases performed best, and left HS cases showed an intermediate level of performance (see also Table 2 for summary of all SLT metrics). Formal comparisons focused on “SLT Learning” (Trial 1–3 sum) as the primary outcome, with “Learning Efficiency” examined as a secondary confirmatory index.

FIGURE 1.

SLT performance across HS groups. Bilateral HS showed the lowest performance overall, with significantly reduced “SLT Learning” (A) and “Learning Efficiency” (B) compared to right HS. “SLT Peak” (Highest Score; C) significantly differed across groups, but no pairwise contrasts survived correction. “SLT Delay” (D) was lowest in bilateral HS, although group differences were not statistically significant. BHS, bilateral hippocampal sclerosis; LHS, left hippocampal sclerosis; RHS, right hippocampal sclerosis; SLT, spatial learning task.

TABLE 2.

SLT summary scores by group.

Variable	N	Groups			Comparison
		Bilateral HS (N = 16)	Left HS (N = 28)	Right HS (N = 19)	p b	BF10	Contrasts
Trial 1	63	5.0 (3.8, 6.3)	6.0 (4.0, 8.0)	8.0 (6.0, 9.0)	0.02	3.5	BHS < RHS
Trial 2	63	5.0 (3.5, 8.0)	8.0 (6.0, 9.0)	9.0 (8.0, 9.0)	<0.001	257	BHS < LHS BHS < RHS LHS < RHS
Trial 3	48	8.0 (4.5, 9.0)	8.0 (6.0, 9.0)	9.0 (6.0, 9.0)	0.5	0.4	–
Trial 4	30	6.0 (5.0, 7.8)	9.0 (5.8, 9.0)	6.0 (5.8, 6.5)	0.2	0.4	–
Trial 5	12	6.0 (5.3, 6.8)	9.0 (6.0, 9.0)	7.0 (7.0, 7.5)	0.4	0.6	–
Highest score	63	8.5 (5.0, 9.0)	9.0 (9.0, 9.0)	9.0 (9.0, 9.0)	0.02	6.5	–
SLT learning a	63	17.5 (11.0, 23.3)	22.0 (18.0, 26.0)	25.0 (23.5, 27.0)	0.004	17.8	BHS < RHS
SLT delay	53	8.0 (5.0, 9.0)	9.0 (8.0, 9.0)	9.0 (8.0, 9.0)	0.08	1.4	–
Learning efficiency	63	0.7 (0.5, 0.9)	0.8 (0.7, 1.0)	0.9 (0.9, 1.0)	0.003	23.7	BHS < RHS
Learning class	63				0.04	6.2	RHS↑ Ceiling; RHS↓ Limited Improvers; BHS↑ Limited Improvers
Ceiling		3.0 (18.8%)	10.0 (35.7%)	11.0 (57.9%)
Improvers		8.0 (50.0%)	14.0 (50.0%)	8.0 (42.1%)
Limited Improvers		5.0 (31.3%)	4.0 (14.3%)	0.0 (0.0%)

Note: Scores beyond Trial 5 were not presented due to the small number of patients.

Abbreviations: BF, Bayes factor; HS, hippocampal sclerosis; IQ, intelligence quotient; TLE, temporal lobe epilepsy.

^a SLT Learning computed by summing Trial 1 to 3, with scores carried forward for patients with missing values on Trial 3.

^b Kruskal–Wallis test for all continuous SLT indices; χ ² test for learning class.

3.2.1. Primary outcome: SLT learning

“SLT Learning” significantly differed across HS groups, with strong evidence for group differences (χ ²[2] = 10.95, p = 0.004, η ²[H] = 0.15, BF₁₀ = 17.76; large effect size [ES]). Post hoc Wilcoxon tests with Holm correction revealed that participants with bilateral HS scored significantly lower than those with right HS (p = 0.005, BF₁₀ = 25.56, r = 0.53; large ES). Differences between bilateral and left HS (p = 0.09, BF₁₀ = 1.41, r = 0.28; small ES) and between left and right HS (p = 0.09, BF₁₀ = 2.08, r = 0.30; small ES) were not statistically significant.

3.2.2. Secondary and descriptive indices

“Learning Efficiency” showed a consistent pattern, with significant group differences (χ ²[2] = 11.67, p = 0.003, η ²[H] = 0.16, BF₁₀ = 23.68; large ES). Similarly, post hoc Wilcoxon comparisons indicated that only the bilateral HS group scored significantly lower than right HS (p = 0.003, BF₁₀ = 30.48, r = 0.56; large ES). Together, these results suggest that bilateral HS is characterized by reduced learning and learning efficiency on the SLT, with the clearest deficits evident when compared to right HS.

“SLT Peak” (highest score achieved across all learning trials) showed a significant omnibus effect (χ ²[2] = 8.08, p = 0.018, η ²[H] = 0.10; BF₁₀ = 6.45; moderate ES), but no pairwise comparison survived correction for multiple comparisons, suggesting that differences in maximum learning attainment were modest and not clearly lateralized (Figure 1C). “SLT Delay” scores were available for 53 of 63 participants (84%), with non‐completers showing lower peak learning scores than completers (see Supplementary Material B), indicating that attrition disproportionately affected weaker learners. While bilateral HS showed the lowest delayed recall scores nominally (Figure 1D), the extent of missing data prevents firm conclusions regarding group differences in delayed recall.

3.2.3. Covariate analysis

Associations between “SLT Learning” and demographic and clinical variables were examined using non‐parametric methods. No significant relationships were found for age (p = 0.34, BF₁₀ = 0.28), age of seizure onset (p = 0.55, BF₁₀ = 0.36), epilepsy duration (p = 0.08, BF₁₀ = 0.49), or IQ (p = 0.07, BF₁₀ = 1.11). A significant difference of sex was observed, with females scoring higher than males (p = 0.015, BF₁₀ = 2.91; mean difference = 3.6 points). A model including sex as a covariate confirmed a significant main effect of HS laterality (p = 0.0024) and sex (p = 0.01), with no laterality × sex interaction (p = 0.63). Exploratory sex‐stratified analyses are reported in Supplementary Material C.

3.3. “SLT learning” is a specific but not sensitive marker of bilateral HS

To evaluate the SLT's clinical utility as a behavioral indicator of bitemporal dysfunction, we examined how well SLT scores discriminate bilateral HS from unilateral cases (Figure 2). Diagnostic properties (e.g., sensitivity, specificity) of the “SLT Learning” score (sum of Trials 1–3) were calculated to quantify its accuracy in identifying bilateral involvement. This measure was selected for its simplicity and ease of clinical interpretation.

FIGURE 2.

Receiver Operating Characteristic (ROC) Curve for “SLT Learning” in classifying bilateral HS versus unilateral HS. Acc, accuracy; AUC, area under the curve; CI, confidence interval; NPV, negative predictive value; PPV, positive predictive value; Se, sensitivity; Sp, specificity.

Using a cutoff score of <13 (derived from Youden's J), performance in 16 bilateral versus 47 unilateral HS cases yielded a sensitivity of 44% (95% CI [0.20–0.70]; 7/16 correctly identified) and a specificity of 96% (95% CI [0.85–0.99]; 45/47 correctly excluded). Predictive values at the observed prevalence were PPV = 0.78 (95% CI [0.40–0.97]) and NPV = 0.83 (95% CI [0.71–0.92]), with an overall classification accuracy of 83% (95% CI [0.71–0.91]). The AUC = 0.73 indicated acceptable discrimination. Likelihood ratio analyses indicated that poor SLT performance markedly increased the probability of bilateral hippocampal compromise (LR+ = 10.3, 95% CI [2.4–44.5]), whereas intact performance only modestly reduced it (LR− = 0.59, 95% CI [0.40–0.90]).

To further quantify this association, a binary logistic regression was conducted using “SLT Learning” as a continuous predictor. The model was statistically significant (p = 0.005; BF₁₀ = 19.7), indicating that lower scores are systematically associated with a higher likelihood of bilateral HS. Each one‐point decrease is associated with a 15.5% increase in the odds of bilateral HS classification (OR = 0.87, 95% CI [0.78, 0.96], Nagelkerke R ² = 0.20).

3.4. Distribution of learning profiles highlights group differences in acquisition patterns

As a single cut‐off score may not capture the heterogeneity of performance, trial‐by‐trial learning profiles were analyzed to identify acquisition patterns that could differentiate unilateral from bilateral HS. The distribution of learning profiles, derived from the “Relative Gain” metric and classified as Ceiling, Improvers, or Limited Improvers, is shown in Figure 3.

FIGURE 3.

The distribution of learning profiles (Ceiling, Improvers, Limited Improvers) varied across HS laterality groups: Bilateral HS showed the highest proportion of Limited Improvers; right HS was dominated by Ceiling; left HS displayed a more balanced mix. Percentages were calculated separately within each group. BHS, bilateral hippocampal sclerosis; LHS, left hippocampal sclerosis; RHS, right hippocampal sclerosis.

Limited Improvers were most prevalent in the bilateral HS group (31%; 5/16), while no Limited Improvers were observed in the right HS group. Conversely, Ceiling performance was most common among right HS participants (58%; 11/19) and least common in bilateral HS (19%; 3/16). The left HS group exhibited a more balanced distribution across all three profiles.

A chi‐square test indicated a significant association between learning profiles and HS group (χ ²[4] = 9.68, p = 0.046, Cramér's V = 0.28, BF₁₀ = 2.26; moderate ES). Examination of standardized residuals revealed that right HS participants were overrepresented in the Ceiling group (+2.13) and underrepresented in the Limited Improvers group (−2.13). In contrast, bilateral HS participants were overrepresented in the Limited Improvers group (+2.25); and while there was a trend for bilateral HS to be underrepresented in the Ceiling group (−1.84), this did not reach statistical significance. Individual learning trajectories stratified by learning profiles and HS groups are visualized in Supplementary Figure D1.

3.5. Unilateral HS cases with poor SLT performance often show subtle contralateral structural abnormalities

Four patients with MRI‐defined unilateral HS, all on the left, showed poor SLT performance that fell within the “Limited Improver” group, with two patients falling below the cutoff of 13. Upon review of these four cases relative to the remaining unilateral cases, the most notable observation was MRI features of atrophy or signal change (though not collectively meeting a diagnosis of HS) in three of the four patients:

1. Left HS with subtle T2 hyperintensity in the right hippocampus.

2. Severe left HS with right hippocampal atrophy.

3. Left HS with right hippocampal atrophy.

The fourth case showed no right hippocampal abnormality on MRI (Supplementary Figure E1). Notably, apart from one left HS case showing subtle contralateral atrophy, such contralateral hippocampal abnormalities (atrophy or signal change) were not observed in any of the other 42 unilateral HS cases in the “Improvers” or “Ceiling” groups.

4. DISCUSSION

This study examined the utility of the SLT in indexing bitemporal dysfunction in TLE, using bilateral HS as the objective marker of bitemporal compromise against which task performance was validated. The hypothesis that bilateral cases would perform more poorly was partly supported. A Trial 1–3 learning score < 13 was a highly specific marker of bitemporal structural pathology (specificity = 96%), although sensitivity was modest (44%), indicating that higher scores did not reliably exclude bilateral structural compromise. While patients with unilateral HS generally performed well, those with left HS showed slower early learning compared to right HS, though typically reached ceiling with repetition. Notably, four left HS cases were classified as “limited improvers”, two of whom scored below the <13 cutoff; while this group only satisfied MRI‐defined HS criteria unilaterally, three of the four exhibited features of structural compromise in the non‐HS affected hippocampus.

These findings position the SLT as a rule‐in marker of bitemporal structural pathology: scores below 13 were uncommon in unilateral HS, suggesting that most unilateral cases retained sufficient capacity to learn the task. This may reflect the task's strategic flexibility, as it can be solved using both verbal (categorical, such as ‘Chair–Middle’) and nonverbal (coordinate‐based) strategies, and in this sense is not critically dependent on either hippocampus.

The SLT is, however, not a rule‐out marker of bitemporal structural pathology, as many patients with bilateral HS scored ≥13 across three trials. One possibility is that performance may be supported by residual hippocampal integrity or contributions from adjacent medial temporal cortices such as the perirhinal and parahippocampal regions. ¹⁷ While intact scores were more frequent in unilateral HS compared to bilateral HS, they were not uncommon in the latter, indicating that a score ≥13 reduces but does not eliminate the likelihood of bitemporal involvement. Accordingly, intact performance should be weighed as part of a broader neuropsychological and multimodal evaluation (e.g., MRI, PET, EEG, and fMRI) rather than taken as evidence of absence.

Within unilateral HS, asymmetries in learning efficiency were evident. Left HS cases attained lower scores on early trials but typically reached ceiling with further repetition, suggesting that difficulty lay in the rate of acquisition rather than the capacity to learn. Right HS, by contrast, more often performed at ceiling from the outset. This distinction was mirrored in the learner profiles, with left HS overrepresented among “Improvers” (showing >40% of possible gain across trials) and right HS among “Ceiling” performers (scoring 8/9 or 9/9 on Trial 1). One possible strategy‐level explanation is that individuals often default to categorical, verbally mediated coding, which is cognitively efficient, rather than to more effortful nonverbal/coordinate‐based strategies. ¹⁸ , ¹⁹ In left HS, impairment of verbal coding could slow initial acquisition, but with repeated exposure patients may recruit alternative non‐verbal strategies, allowing performance to normalize. An alternative possibility is that patients continue to rely on verbal strategies throughout, but inefficiencies in verbal coding slow learning. Both these interpretations align with the observed pattern of slower starts but intact eventual capacity, though our data cannot directly confirm the underlying mechanisms.

In the analysis of unilateral HS cases with notably poor performance, we observed that patients with MRI‐defined unilateral HS may still show subtle contralateral abnormalities, which in some instances may be sufficient to disrupt associative learning. Among the four left HS cases with “Limited Improver” profiles, three showed subtle contralateral hippocampal changes (atrophy or T2 signal change, though not meeting diagnostic criteria for HS). With the exception of one case, the remaining 42 unilateral HS cases with “ceiling” or “improver” profiles showed no contralateral hippocampal atrophy or signal change. This pattern suggests that markedly poor SLT performance may serve as a behavioral marker of covert bitemporal involvement, but such abnormalities may not invariably impair performance.

4.1. Clinical implications

The SLT is intended to complement, not replace, comprehensive neuropsychological assessment and multimodal investigation. In presurgical evaluation, the SLT Trial‐1–3 sum is best regarded as a probabilistic rule‐in indicator of bitemporal involvement. From a practical standpoint, administration should typically include at least three trials; if ceiling (9/9) is achieved at Trial 2, the Trial‐2 score may be carried forward to Trial‐3, although extension to ~5 or more trials may be helpful where slower early acquisition raises questions as to whether improvement may occur (i.e., additional trials might be needed to clarify whether improvement is ultimately achieved versus consistently poor performance despite repeated exposure). A cumulative T1‐3 score below 13 materially increases concern for bitemporal compromise (OR ≈ 1.16 per point lower) and should prompt expedited corroboration with multimodal methods alongside counseling regarding postoperative amnesia risk. Importantly, the SLT should not be used in isolation; the aim should always be to interpret in conjunction with other memory tests with established lateralising value. ²⁰ Where corroborative evidence of bitemporal dysfunction is strong, resective surgery is contraindicated in all but palliative cases given the risk of profound memory decline. ²¹

Demographic factors had limited influence on SLT performance, consistent with previous work. ¹⁰ Although females scored higher on average, the effect of HS was not moderated by sex (see Supplementary Material C). Neither age nor IQ affected performance. This highlights the potential value of the SLT in populations for whom standard verbal or visuoconstructional tasks (e.g., list learning or complex figure recall) may be less suitable, such as individuals with language impairments, lower literacy, comorbid intellectual disability, or non‐English speaking patients from diverse cultural backgrounds. Contemporary normative data for adults aged 18–45 are available ¹⁰ and should be consulted when interpreting individual trial scores in clinical practice, though these norms do not apply to the composite “SLT Learning” sum. These norms may not be directly applicable to the present dataset given that a proportion of the cohort exceeded this age range; nonetheless, age itself was not significantly associated with SLT performance in this cohort (Section 3.2.3).

4.2. Limitations

The retrospective nature of this study introduces several sources of potential bias. First, there was variability in administration practices across the study period (e.g., variable number of trials, inconsistent collection of delayed recall). This variability meant that the composite “SLT Learning” variable relied on carried‐forward scores for a subset of participants (Section 2.2) and is therefore not a perfect measure of Trial 3 performance. However, the robustness of the primary findings is supported by two entirely imputation‐free analyses—“Learning Efficiency” and the trial‐by‐trial learning profile classification—which showed the same pattern of group differences. Patients who did not complete the delayed recall trial also had lower peak scores than delay completers (see Supplementary Material B), ultimately precluding us from drawing meaningful conclusions about laterality effects on delayed recall scores. Second, potential state factors such as medication load, seizure timing, fatigue, and psychopathology could not be systematically accounted for. It is known that such variables can influence memory performance and may represent genuine sources of unmeasured variance; their potential contribution to the observed group differences cannot be determined from the current data. Additionally, precipitating injury data (including febrile convulsions, head trauma, and central nervous system infection) were not systematically recorded in a format amenable to reliable extraction across the full study period (1990–2024). These factors may contribute to bilateral structural compromise and represent a potentially relevant unmeasured variable. Third, as the data were collected retrospectively from clinically directed assessments (versus a uniform, prospective protocol) additional memory (and non‐memory) tests with lateralising value such as the Rey Auditory Verbal Learning Test and figural memory tasks ²² , ²³ were not administered to all cases. This prevented systematic comparison with other memory measures, and consequently, the SLT's performance could not be contextualized within the broader memory profile of each group. This is an important limitation because the clinical value of the SLT is best understood in conjunction with, not in place of, established measures with lateralising value; as noted above, SLT scores must be interpreted in the context of a broader neuropsychological evaluation and additional multimodal data. Finally, postoperative neuropsychological data were not available for the full sample. From a clinical standpoint, the most consequential question is not whether the SLT detects bilateral structural pathology, but whether it identifies patients at elevated risk of memory deterioration following unilateral resection. Prospective studies tracking pre‐ and postoperative SLT performance alongside standardized memory outcomes would clarify whether the SLT adds incremental predictive value beyond existing presurgical markers.

5. CONCLUSIONS

This study demonstrates that poor performance on the SLT (Trial‐1–3 sum <13) is a highly specific indicator of bitemporal hippocampal pathology, with a low score materially increasing the probability of bilateral involvement. Intact scores, however, do not reliably exclude bilateral structural compromise. Left‐sided hippocampal structural compromise was associated with slower initial learning, but patients typically caught up with repetition, indicating reduced efficiency rather than reduced capacity. These findings highlight the utility of the SLT as an elementary, easy‐to‐administer task that can inform presurgical evaluation, particularly in populations for whom standard neuropsychological measures may be less suitable. A low SLT score should prompt targeted review of multimodal evidence for bitemporal involvement, given the associated elevation in surgical memory risk.

AUTHOR CONTRIBUTIONS

Andy Sitoh: Conceptualization; project administration (lead); formal analysis (lead); and writing—original draft (lead). David Weintrob: Conceptualization; supervision (equal); resources (equal); and writing—review and editing (equal). Jacqueline F. I. Anderson: Supervision (equal) and writing—review and editing (equal). Marie F. O'Shea: Resources (equal) and writing—review and editing (supporting). Graeme D. Jackson: resources (equal) and writing—review and editing (supporting). Piero Perucca: Resources (equal) and writing—review and editing (supporting). Chris Tailby: Conceptualization (lead); project administration; supervision (lead); and writing—review and editing (lead).

CONFLICT OF INTEREST STATEMENT

PP is supported by an Emerging Leadership Investigator Grant from the Australian National Health and Medical Research Council (APP2017651), the University of Melbourne, Monash University, the Austin Medical Research Foundation, and the Norman Beischer Medical Research Foundation. He has received speaker honoraria or consultancy fees to his institution from Chiesi, Eisai, LivaNova, Novartis, Sun Pharma, Supernus, and UCB Pharma, outside the submitted work. He is an Associate Editor for Epilepsia Open. The remaining authors have no conflicts of interest. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

Supporting information

Supplementary Material A. Sensitivity Analysis Restricted to Unilateral HS Cases with Unambiguous Lateralization

Table A1. Sample characteristics by group with unambiguous lateralisation.

Supplementary Material B. Delay Completers vs. Non‐Completers

Table B1. Comparison between SLT delay completers versus non‐completers.

Supplementary Material C. Sex‐Adjusted ROC Analysis and Thresholds

Table C1. Sex‐stratified diagnostic accuracy of SLT learning for classifiying bilateral HS versus unilateral HS.

Figure D1. Individual learning trajectories across HS groups, with each panel plotting trial‐by‐trial performance. Bilateral HS showed a higher proportion of Limited Improvers, with flatter curves and less consistent gains. Right HS was dominated by Ceiling performers who reached maximum scores within the first trial. Left HS displayed a mix of all three profiles, reflecting greater variability in learning patterns.

Figure E1. MRI images for four patients with unilateral left HS with poor SLT performance.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.