Naturalistic Language-related Movie-Watching fMRI Task for Detecting Neurocognitive Decline and Disorder

Helen Meng; Hoi-lam Helene Fung; Man Wai Mak; Patrick Wong; Xianmin Gong; Xixin Wu; Yuejiao Wang

REVIEW 4 major objections 5 minor 40 references

Naturalistic Language-related Movie-Watching fMRI Task for Detecting Neurocognitive Decline and Disorder

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Passive movie-watching fMRI can separate older adults with cognitive decline from those without it, with an average AUC of 0.86.

desk verdict A novel naturalistic language fMRI task with a plausible motivation, but the 0.86 AUC is an overfit estimate; the study needs a proper nested evaluation before the claim can be taken seriously. read the letter →

arxiv 2506.08986 v1 pith:QKBQ4QFU submitted 2025-06-10 cs.CL

Yuejiao Wang , Xianmin Gong , Xixin Wu , Patrick Wong , Hoi-lam Helene Fung , Man Wai Mak , Helen Meng This is my paper

classification cs.CL

keywords naturalisticfMRItaskmovie-watchingneurocognitivedisorderdetectioncognitivedeclinescreeninglanguagenetworkMontrealAssessmentspeechversussilencecontrastmachinelearningclassification

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that cognitive decline can be detected from brain activity while an older adult simply watches a short movie in an MRI scanner, with no explicit task to perform. The movie is a naturalistic Cantonese family story, and the fMRI contrast of interest is brain response during speech segments versus silence segments. On 97 non-demented community-dwelling older adults, classifiers using these speech-related brain maps together with age, gender, and education reached an average area under the curve (AUC) of 0.86 for separating participants scoring at or below 20 on the Montreal Cognitive Assessment (MoCA) from those scoring above 20. If this result holds, it would make early screening for neurocognitive disorder possible in a passive, ecologically valid setting.

What carries the argument

The central object is the per-participant speech-versus-silence T-map. Each voxel's fMRI time series is regressed on speech-event and silence-event regressors convolved with the canonical hemodynamic response function, and the contrasts $\beta_1-\beta_2$ and $\beta_2-\beta_1$ are converted to t-statistics, giving two whole-brain maps: one for activation by speech and one for deactivation by speech. Dimension reduction keeps only voxels whose T-values correlate with MoCA at $p<0.01$ in the training fold, then applies L1-penalized linear support vector machine feature selection; the surviving features, concatenated with demographics, go into a support vector classifier or a Gaussian naive Bayes classifier. Evaluation is 500 iterations of stratified shuffle-split cross-validation, with performance reported as AUC and feature-set differences tested by the Wilcoxon signed-rank test.

What would settle it

Run the same pipeline on an independent cohort of several hundred older adults with the voxel mask fixed from the original 97 participants and measure AUC for the same NORMAL versus DECLINE split; if the AUC falls toward chance, the reported 0.86 came from per-fold feature selection rather than from the movie task's signal. A simpler check is to move the MoCA cutoff from 20 to nearby thresholds and ask whether classification stays above chance outside the exact median split.

Watch

Extended reading notes

Core claim

Passively watching a naturalistic movie produces whole-brain speech-versus-silence T-maps that separate cognitively normal older adults from those with measurable decline. The separation is not explained by demographics alone: adding the T-maps to age, gender, and education significantly raises AUC, from about 0.73 to 0.76 for demographics only up to 0.862 with all features in the support vector classifier. The features the data-driven selection keeps returning to are language-related regions---superior temporal gyrus, middle temporal gyrus, and right cerebellum---while the complementary silence-versus-speech map covers posterior cingulate and precuneus areas associated with the default mode network. The paper reads this as evidence that language-related fMRI from a naturalistic task can detect cognitive decline before overt neurocognitive disorder symptoms appear.

Load-bearing premise

The load-bearing premise is that splitting Montreal Cognitive Assessment scores at the median, 20, creates two stable cognitive groups, and that the training-set step of selecting voxels by their correlation with MoCA generalizes beyond the roughly five participants held out in each cross-validation fold.

Editorial extensions

If this is right

A passive movie-watching scan could serve as an early screening step for cognitive decline in older adults, including people who have trouble following traditional neuropsychological tests.
Speech-versus-silence brain maps supply information beyond age, gender, and education, so brain language responses are measuring decline directly rather than proxying for demographics.
Including both speech-activated and speech-deactivated regions improves classification, meaning the default-mode-network-like suppression captured by the silence-versus-speech map is also informative.
The most informative voxels residing in temporal language areas and the cerebellum supports the paper's claim that naturalistic language processing is a sensitive early marker of neurocognitive decline.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Editorial extension: because the feature-selection step already correlates voxel T-values with continuous MoCA, a direct regression of MoCA score on brain features may be more stable than the binary 20-point median split and is worth testing.
Editorial extension: a head-to-head comparison of this naturalistic task with resting-state fMRI or a conventional picture-naming task in the same participants would isolate how much ecological validity contributes to detection accuracy.
Editorial extension: the strongest test would be longitudinal---whether the T-map features predict conversion to clinical neurocognitive disorder or dementia within a few years, something the paper leaves to future work.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

Here's my take on arXiv:2506.08986. The idea is genuinely new: a passive, naturalistic movie-watching fMRI task in Cantonese, with a speech-versus-silence contrast, used to classify cognitive decline defined by MoCA. That combination isn't in the literature they cite, and the task has real ecological validity for older adults. The overlap analysis with the LanA language atlas is a nice sanity check. The writing is clear and the authors acknowledge limitations, including the lack of clinical diagnosis.

The problem is the evaluation. The 0.86 AUC is not a reliable estimate of how the task would perform out-of-sample. Three specific issues stand out. First, the T-map mask is computed across all participants before the train/test split, so the feature space itself is constructed with test-set information. Second, the feature-selection step keeps voxels whose T-values are Pearson-correlated with the same MoCA scores that define the labels. That is not inherently circular if done inside the training fold, but with a median split at MoCA 20 and a sample of 97, the selected voxels are likely fitting idiosyncratic training-set correlations. Third, each test fold has about five participants. Averaging 500 overlapping shuffle-split iterations does not fix the fact that each AUC is a coarse discrete statistic, and the standard deviations (around 0.2) show how unstable the estimates are. There is no permutation test and no nested cross-validation, so the null distribution is unknown; a pipeline like this could plausibly produce high AUCs by chance on this sample size.

The demographics-only model already reaches AUC 0.73-0.755, so the incremental value of the fMRI features is the key claim. That claim needs a much more rigorous evaluation. The authors should compute the mask inside the training loop, use nested CV or leave-one-subject-out, run a permutation null, and ideally validate on an independent sample with clinical diagnosis. Without that, the paper is a proof-of-concept, not a demonstrated biomarker. The stress-test note holds up on reading.

For a conference track, I'd send it to review because the task is worth knowing about and the flaws are fixable, but I'd expect major revision. For your own work, I wouldn't cite the 0.86 figure as evidence.

Referee Report

4 major / 5 minor

Summary. The paper proposes a naturalistic Cantonese movie-watching fMRI task as a tool for detecting cognitive decline in older adults. Ninety-seven non-demented Hong Kong Chinese older adults watched an 11-minute movie during fMRI; individual speech-vs-silence and silence-vs-speech t-maps were computed with a GLM. Participants were labeled NORMAL (MoCA > 20) or DECLINE (MoCA ≤ 20) using a median split of the Hong Kong MoCA. After a two-step feature selection (Pearson correlation with MoCA followed by L1-penalized SVC selection) applied to sigma-masked t-maps, features were combined with demographics and fed to an SVC or Gaussian Naive Bayes classifier. The authors report an average AUC of 0.86 over 500 stratified shuffle-split iterations, with selected features localized to language-related regions (MTG, STG, cerebellum). The central claim is that this passive movie-watching fMRI task, combined with demographics, can detect cognitive decline and potentially early neurocognitive disorder.

Significance. If the claimed AUC of 0.86 were obtained under a valid evaluation protocol, the study would provide a valuable ecologically valid fMRI task for cognitive screening, with the advantage of minimal participant burden. The data collection itself (97 older adults with naturalistic movie watching) is a useful resource, and the overlap of the derived speech t-maps with the language network atlas (LanA) is a credible sanity check. The paper also deserves credit for making the feature-localization results interpretable by linking selected voxels to established language areas. However, the evaluation protocol has several load-bearing methodological problems that currently prevent the reported performance from being accepted as evidence for the task's utility.

major comments (4)

[Section 3.2] The voxel masking step is performed 'across all participants' before the train/test split. Specifically, the masked T-maps are defined by keeping voxels with T values above a significance threshold using data from all 97 participants, and only then is the dataset split into training and test folds. This leaks test-set information into the feature space construction. Even if the subsequent two-step feature selection is refit on training folds, the set of candidate voxels has already been determined using held-out participants. This can inflate the reported AUC and should be corrected by deriving the mask inside each training fold only, or by an outer cross-validation loop that includes the masking step.
[Section 3.2 and Section 4.2] The test set in each stratified shuffle-split contains only about 5 participants (95% training / 5% test of 97). Each per-iteration AUC is therefore a coarse statistic with very high variance, and the 500 iterations are not independent because the splits overlap heavily. Averaging these overlapping AUCs and conducting Wilcoxon signed-rank tests on the 500 values (Section 4.2) does not provide a valid statistical comparison. Moreover, no permutation test or nested cross-validation is reported, so the null distribution of the 0.86 AUC is unknown. A permutation test that shuffles the NORMAL/DECLINE labels and runs the entire pipeline (including masking and feature selection) is essential to establish whether the observed AUC exceeds chance.
[Section 3.2 and Section 5] The binary labels are a median split of MoCA at 20. This cutpoint is arbitrary and does not correspond to a clinical diagnosis of neurocognitive disorder; the paper acknowledges this limitation in the final paragraph, yet the abstract and conclusions still claim the task detects 'cognitive decline and early NCD.' The reported AUC only measures separability of two halves of a screening-test score distribution in a non-demented sample. Additional analyses are needed to support the claims: for example, treating MoCA as a continuous outcome, testing sensitivity to the choice of threshold, or validating against an independent clinical or neuropsychological criterion. Without such evidence, the conclusions overstate the clinical implication of the result.
[Section 4.2 and Table 2] The standard deviations of the AUC distributions are large (e.g., 0.178 for the best SVC model). With a test-set size of about 5, individual AUC values can vary from 0 to 1 in a single split, and the reported mean of 0.862 over overlapping splits does not indicate stable model performance. The paper should report confidence intervals based on independent resampling (e.g., repeated non-overlapping k-fold or bootstrap with feature selection repeated in each resample) and should also report the distribution of per-subject predictions to make clear how many participants are actually correctly classified.

minor comments (5)

[Abstract] The word 'neurcognitive' appears in the abstract; this appears to be a typo for 'neurocognitive'.
[Section 3.2] The phrase 'T values above the statistical significance threshold' does not state the threshold value or whether it is corrected for multiple comparisons. Also, 'M oCA' has an extra space.
[Figure 2] The scissors icon indicating 'the feature is optional' is not explained in the caption; please clarify which features are optional and how the reader should interpret the diagram.
[Section 3.1] The GLM equation describes contrasts but does not explicitly state that the t-values are computed as C_i divided by the pooled standard error; this is stated in the text, but a more formal definition in the equation or surrounding text would improve reproducibility.
[Section 4.3] The localization results are based on selection frequency across 500 overlapping splits, but because the splits are not independent, the reported frequencies are not interpretable as stable probabilities. A more rigorous assessment would use a nested cross-validation or bootstrap procedure that accounts for the dependence.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the MoCA-driven feature selection is performed inside each training fold, and the held-out AUC is a genuine generalization estimate, though it carries data-leakage caveats outside the circularity definition.

full rationale

The paper's central claim is an empirical classification result, not a derived first-principles identity. Labels are defined by thresholding MoCA, and the two-step feature selection (Pearson correlation with MoCA, then L1-penalized SVC) is applied to the training fold in each stratified shuffle-split iteration; the classifier is then evaluated on held-out participants whose labels were not used in feature selection or model fitting, so the reported average AUC of 0.86 is not forced to equal the training labels by construction. The only design choice that touches test information is the pre-split 'masked T maps' threshold computed across all participants, which is a data-leakage/optimism concern rather than a circular reduction: the mask does not encode individual test labels and the final classifier weights are still estimated from training data. No self-citations, imported uniqueness theorems, or ansatz-by-citation chains are load-bearing, and the LanA atlas provides an external, independent benchmark for the language-relatedness of the T-maps. The authors also explicitly state that future work is needed for clinical NCD diagnoses, acknowledging the current result is a screening-level association. Under the required standard (exhibit a specific reduction, e.g., Eq. X = Eq. Y by construction), no circular step is present.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The study's central result depends on several assumptions: the validity of the MoCA median split as a proxy for cognitive decline, the correctness of the GLM and HRF modeling, and the arbitrary thresholds in the feature-selection pipeline. No new physical entities are introduced. The free parameters and arbitrary thresholds collectively account for the performance and reduce the independent evidentiary value of the reported AUC.

free parameters (5)

MoCA threshold for group assignment = 20 (sample median)
Used to define NORMAL and DECLINE labels in Section 3.2; chosen post hoc as the median of the observed sample.
Pearson correlation p-value threshold for feature selection = 0.01
Used in step-1 feature selection in Section 3.2; arbitrarily chosen, no justification.
T-map masking significance threshold = Not reported
Required to create masked T-maps in Section 3.2; the exact statistical threshold is omitted, so it is a free parameter set by the authors.
L1 feature selection threshold = 1e-5 (sklearn default)
Used in step-2 feature selection via SelectFromModel in Section 3.2; left at the library default.
SVC regularization parameter C = 1.0 (sklearn default)
Used for the SVC classifier in Section 3.2; not tuned.

assumptions (4)

domain assumption The canonical hemodynamic response function (HRF) accurately models the BOLD response in older adults.
Used in the GLM in Section 3.1; if the HRF is misspecified for the aging population, the T-maps are biased.
domain assumption The speech and silence event annotations of the movie 'Sweet Home' are accurate and capture language-related processing.
The contrasts C1 and C2 in Section 3.1 rely on these annotations; incorrect annotations would mislabel the neural conditions.
ad hoc to paper A median split of MoCA at 20 creates two meaningful cognitive groups (NORMAL and DECLINE) in this sample.
The binary label in Section 3.2 depends on this split; MoCA is a screening test, not a clinical diagnosis, and the median split is sample-derived.
standard math Standard SPM12 preprocessing and first-level GLM produce unbiased activation maps.
Assumed in Section 2 for fMRI data preprocessing and in Section 3.1 for T-map generation.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Naturalistic Language-related Movie-Watching fMRI Task for Detecting Neurocognitive Decline and Disorder." pith.science (2026). https://pith.science/paper/QKBQ4QFU

@misc{pith2026250608986,
  author       = {Pith},
  title        = {Pith review of: Naturalistic Language-related Movie-Watching fMRI Task for Detecting Neurocognitive Decline and Disorder},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QKBQ4QFU}},
  note         = {Machine review of arXiv:2506.08986}
}

read the original abstract

Early detection is crucial for timely intervention aimed at preventing and slowing the progression of neurocognitive disorder (NCD), a common and significant health problem among the aging population. Recent evidence has suggested that language-related functional magnetic resonance imaging (fMRI) may be a promising approach for detecting cognitive decline and early NCD. In this paper, we proposed a novel, naturalistic language-related fMRI task for this purpose. We examined the effectiveness of this task among 97 non-demented Chinese older adults from Hong Kong. The results showed that machine-learning classification models based on fMRI features extracted from the task and demographics (age, gender, and education year) achieved an average area under the curve of 0.86 when classifying participants' cognitive status (labeled as NORMAL vs DECLINE based on their scores on a standard neurcognitive test). Feature localization revealed that the fMRI features most frequently selected by the data-driven approach came primarily from brain regions associated with language processing, such as the superior temporal gyrus, middle temporal gyrus, and right cerebellum. The study demonstrated the potential of the naturalistic language-related fMRI task for early detection of aging-related cognitive decline and NCD.

Figures

Figures reproduced from arXiv: 2506.08986 by the authors.

**Figure 1.** The proposed movie-watching fMRI task. While being scanned in the MRI scanner, the participant watches a movie projected onto a mirror placed inside the scanner. The participant receives the movie’s audio stream through headphones. out the names of a series of objects) used for screening and detecting NCD are typically simple and focus on basic language functions at the word or verbal level [9, 10]. Additionally, … view at source ↗

**Figure 2.** The flowchart of machine learning-based cognitive status classification. T-maps obtained from the movie-watching fMRI task underwent a two-step dimension reduction process and were concatenated with demographics for cognitive status classification. Scissors means the feature is optional. and spatial smoothing with an isotropic 5 mm full-width-athalf-maximum Gaussian kernel. MoCA test for cognitive status. As this s… view at source ↗

**Figure 3.** (a) An example Speech vs Silence T-map in three different view perspectives (from a female subject with age = 63 and MoCA score = 24). (b) The probabilistic map showing the probability (out of 500 times of experiments) of each voxel being selected after the step-1 feature selection, with green markers representing the top three most frequently selected voxel clusters after the step-2 feature selection. (c) The distr… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

40 extracted references · 40 canonical work pages

[1]

Naturalistic Language-related Movie-Watching fMRI Task for Detecting Neurocognitive Decline and Disorder

Introduction The growing aging population poses a significant challenge to society. In China, older adults aged 60 and above constitute 18.70% of the total population [1]. Aging typically leads to neu- rocognitive decline, which may progress into neurocognitive disorder (NCD) that is more severe and goes beyond the normal cognitive aging trajectory. Globa...

work page Pith review arXiv 2025
[2]

Sweet Home

Data Collection Participants. 97 Chinese older adults aged above 60 years (see Table 1 for more demographic information) residing in Hong Kong were recruited. All participants were right-handed, had normal or correct-normal visual acuity, could read Chinese, without dementia, and were eligible for fMRI scanning. Eth- ical approval was obtained from the me...

work page
[3]

Approach 3.1. Language-related Statistical T-maps Extraction For each participant, we used general linear modeling to analyze their brain activation related to speech processing: Y = β0 + β1(Speech event∗ HRF ) +β2(Silence event∗ HRF ) +Headmotion Regressors+ Residual (1) where the fMRI signals Y (a 1-D time series) within each brain voxel of each partici...

work page
[4]

Results and Discussions 4.1. T-maps’ Coverage of Language-related Brain Areas To demonstrate that the Speech vs Silence T-maps extracted from the movie-watching task (see Figure 3a) captured the ac- tivity of brain regions associated with language processing, we compared the distribution of voxels in the masked Speech vs Silence T-map against a validated ...

work page
[5]

the Speech vs Silence T-map captured brain activity associ- ated with language processing. 4.2. Results of Cognitive Status Classification The average AUC of 500 stratified shuffle-split iterations is shown in Table 2. As expected, the performance of models based on demographic features was not bad, due to the influence of age and education on cognitive a...

work page
[6]

Conclusions In this paper, we introduced and demonstrated the effectiveness of a naturalistic language-related movie-watching fMRI task for detecting cognitive decline and early NCD. The language- related fMRI features (statistical T-maps) extracted from the task, combined with demographics, achieved an averaged AUC of 0.86 for classifying participants’ c...

work page
[7]

T45-407/19N)

Acknowledgements This research is partially supported by the HKSARG Re- search Grants Council’s Theme-based Research Grant Scheme (Project No. T45-407/19N)

work page
[8]

Bulletin of the seventh national census (no. 5),

N. B. of Statistics of China, “Bulletin of the seventh national census (no. 5),” 2023. [Online]. Available: https: //www.stats.gov.cn/sj/zxfb/202302/t20230203 1901085.html

work page 2023

Show all 40 references

[9]

Dementia,

W. H. Organization, “Dementia,” 2023. [Online]. Available: https://www.who.int/news-room/fact-sheets/detail/dementia

2023
[10]

Status and development of care services for the elderly with cognitive impairment,

C. N. C. on Ageings, “Status and development of care services for the elderly with cognitive impairment,” 2021. [Online]. Available: https://mp.weixin.qq.com/s/DSCwFE12Dc9lO3GyQopQ7Q

2021
[11]

Diagnosis of Early Alzheimer’s Disease: Clini- cal Practice in 2021,

A. P. Porsteinsson, R. S. Isaacson, S. Knox, M. N. Sabbagh, and I. Rubino, “Diagnosis of Early Alzheimer’s Disease: Clini- cal Practice in 2021,” The Journal of Prevention of Alzheimer’s Disease, vol. 8, no. 3, pp. 371–386, Jul. 2021

2021
[12]

Multimodal deep learning models for early detection of Alzheimer’s disease stage,

J. Venugopalan, L. Tong, H. R. Hassanzadeh, and M. D. Wang, “Multimodal deep learning models for early detection of Alzheimer’s disease stage,” Scientific Reports, vol. 11, no. 1, p. 3254, Feb. 2021, publisher: Nature Publishing Group

2021
[13]

Early Detection of Alzheimer’s Disease Using Magnetic Res- onance Imaging: A Novel Approach Combining Convolutional Neural Networks and Ensemble Learning,

D. Pan, A. Zeng, L. Jia, Y . Huang, T. Frizzell, and X. Song, “Early Detection of Alzheimer’s Disease Using Magnetic Res- onance Imaging: A Novel Approach Combining Convolutional Neural Networks and Ensemble Learning,” Frontiers in Neuro- science, vol. 14, May 2020

2020
[14]

Cognition, language, and clinical pathological features of non- alzheimer’s dementias: an overview,

J. Reilly, A. D. Rodriguez, M. Lamy, and J. Neils-Strunjas, “Cognition, language, and clinical pathological features of non- alzheimer’s dementias: an overview,” Journal of communication disorders, vol. 43, no. 5, pp. 438–452, 2010

2010
[15]

Aided diagnosis of dementia type through computer-based analysis of spontaneous speech,

W. Jarrold, B. Peintner, D. Wilkins, D. Vergryi, C. Richey, M. L. Gorno-Tempini, and J. Ogar, “Aided diagnosis of dementia type through computer-based analysis of spontaneous speech,” in Pro- ceedings of the Workshop on Computational Linguistics and Clin- ical Psychology: From...

2014
[16]

A review of screening tests for cognitive impairment,

B. Cullen, B. O’Neill, J. J. Evans, R. F. Coen, and B. A. Lawlor, “A review of screening tests for cognitive impairment,”Journal of Neurology, Neurosurgery & Psychiatry, vol. 78, no. 8, pp. 790– 799, 2007

2007
[17]

Screening for mild cognitive impairment: a systematic review,

J. A. Lonie, K. M. Tierney, and K. P. Ebmeier, “Screening for mild cognitive impairment: a systematic review,”International Journal of Geriatric Psychiatry: A journal of the psychiatry of late life and allied sciences, vol. 24, no. 9, pp. 902–915, 2009

2009
[18]

Measures of everyday competence in older adults with cognitive impairment: a systematic review,

L. L. Law, F. Barnett, M. K. Yau, and M. A. Gray, “Measures of everyday competence in older adults with cognitive impairment: a systematic review,” Age and Ageing , vol. 41, no. 1, pp. 9–16, 2012

2012
[19]

Cognitive performance in adults’ daily lives: Is there a lab-life gap?

A. A. Bielak, C. R. Hatt, and M. Diehl, “Cognitive performance in adults’ daily lives: Is there a lab-life gap?” Research in Human Development, vol. 14, no. 3, pp. 219–233, 2017

2017
[20]

Cognition and lan- guage in dementia patients: Contributions from behavior analy- sis,

N. M. Aggio, M. Ducatti, and J. C. de Rose, “Cognition and lan- guage in dementia patients: Contributions from behavior analy- sis,” Behavioral Interventions, vol. 33, no. 3, pp. 322–335, 2018

2018
[21]

Language and dementia: Neuropsy- chological aspects,

D. Kempler and M. Goral, “Language and dementia: Neuropsy- chological aspects,” Annual review of applied linguistics, vol. 28, pp. 73–90, 2008

2008
[22]

Functional mri of language: new approaches to understanding the cortical organization of semantic processing,

S. Bookheimer, “Functional mri of language: new approaches to understanding the cortical organization of semantic processing,” Annual review of neuroscience, vol. 25, no. 1, pp. 151–188, 2002

2002
[23]

A review and synthesis of the first 20 years of pet and fmri studies of heard speech, spoken language and reading,

C. J. Price, “A review and synthesis of the first 20 years of pet and fmri studies of heard speech, spoken language and reading,” Neuroimage, vol. 62, no. 2, pp. 816–847, 2012

2012
[24]

The montreal cognitive assessment, moca: a brief screening tool for mild cognitive impairment,

Z. S. Nasreddine, N. A. Phillips, V . B ´edirian, S. Charbonneau, V . Whitehead, I. Collin, J. L. Cummings, and H. Chertkow, “The montreal cognitive assessment, moca: a brief screening tool for mild cognitive impairment,” Journal of the American Geriatrics Society, vol. 53, no...

2005
[25]

An empirical comparison of spm preprocessing parameters to the analysis of fmri data,

V . Della-Maggiore, W. Chau, P. R. Peres-Neto, and A. R. McIn- tosh, “An empirical comparison of spm preprocessing parameters to the analysis of fmri data,” Neuroimage, vol. 17, no. 1, pp. 19– 28, 2002

2002
[26]

The validity, reliability and clinical utility of the hong kong montreal cognitive assessment (hk-moca) in patients with cerebral small vessel disease,

A. Wong, Y . Y . Xiong, P. W. Kwan, A. Y . Chan, W. W. Lam, K. Wang, W. C. Chu, D. L. Nyenhuis, Z. Nasreddine, L. K. Wong et al., “The validity, reliability and clinical utility of the hong kong montreal cognitive assessment (hk-moca) in patients with cerebral small vessel dis...

2009
[27]

Model- ing the hemodynamic response to brain activation,

R. B. Buxton, K. Uluda ˘g, D. J. Dubowitz, and T. T. Liu, “Model- ing the hemodynamic response to brain activation,” Neuroimage, vol. 23, pp. S220–S233, 2004

2004
[28]

Feature selec- tion based on the svm weight vector for classification of demen- tia,

E. E. Bron, M. Smits, W. J. Niessen, and S. Klein, “Feature selec- tion based on the svm weight vector for classification of demen- tia,” IEEE journal of biomedical and health informatics , vol. 19, no. 5, pp. 1617–1626, 2015

2015
[29]

Spurious but systematic correlations in functional con- nectivity mri networks arise from subject motion,

J. D. Power, K. A. Barnes, A. Z. Snyder, B. L. Schlaggar, and S. E. Petersen, “Spurious but systematic correlations in functional con- nectivity mri networks arise from subject motion,” Neuroimage, vol. 59, no. 3, pp. 2142–2154, 2012

2012
[30]

Population estimate of people with clinical alzheimer’s disease and mild cognitive impairment in the united states (2020–2060),

K. B. Rajan, J. Weuve, L. L. Barnes, E. A. McAninch, R. S. Wil- son, and D. A. Evans, “Population estimate of people with clinical alzheimer’s disease and mild cognitive impairment in the united states (2020–2060),” Alzheimer’s & dementia, vol. 17, no. 12, pp. 1966–1975, 2021

2020
[31]

Wilcoxon signed-rank test,

R. F. Woolson, “Wilcoxon signed-rank test,” Encyclopedia of Bio- statistics, vol. 8, 2005

2005
[32]

Probabilistic atlas for the language network based on pre- cision fmri data from¿ 800 individuals,

B. Lipkin, G. Tuckute, J. Affourtit, H. Small, Z. Mineroff, H. Kean, O. Jouravlev, L. Rakocevic, B. Pritchett, M. Siegelman et al., “Probabilistic atlas for the language network based on pre- cision fmri data from¿ 800 individuals,” Scientific Data, vol. 9, no. 1, p. 529, 2022

2022
[33]

The brain’s default mode network,

M. E. Raichle, “The brain’s default mode network,”Annual review of neuroscience, vol. 38, pp. 433–447, 2015

2015
[34]

Exploring the role of the pos- terior middle temporal gyrus in semantic cognition: Integration of anterior temporal lobe with executive processes,

J. Davey, H. E. Thompson, G. Hallam, T. Karapanagiotidis, C. Murphy, I. De Caso, K. Krieger-Redwood, B. C. Bernhardt, J. Smallwood, and E. Jefferies, “Exploring the role of the pos- terior middle temporal gyrus in semantic cognition: Integration of anterior temporal lobe with ...

2016
[35]

Current controversies on wernicke’s area and its role in language,

J. R. Binder, “Current controversies on wernicke’s area and its role in language,” Current neurology and neuroscience reports , vol. 17, pp. 1–10, 2017

2017
[36]

The cerebellum and language: the story so far,

H. J. De Smet, H. Baillieux, P. P. De Deyn, P. Mari ¨en, and P. Paquier, “The cerebellum and language: the story so far,”Folia Phoniatrica et Logopaedica, vol. 59, no. 4, pp. 165–170, 2007

2007
[37]

The cerebellum and language: historical per- spective and review,

B. E. Murdoch, “The cerebellum and language: historical per- spective and review,”Cortex, vol. 46, no. 7, pp. 858–868, 2010

2010
[38]

Semantic memory and language dysfunction in early alzheimer’s disease: a review,

M. Verma and R. J. Howard, “Semantic memory and language dysfunction in early alzheimer’s disease: a review,” International journal of geriatric psychiatry , vol. 27, no. 12, pp. 1209–1217, 2012

2012
[39]

Syntactic comprehension deficits in alzheimer’s disease,

C. Bickel, J. Pantel, K. Eysenbach, and J. Schr ¨oder, “Syntactic comprehension deficits in alzheimer’s disease,” Brain and Lan- guage, vol. 71, no. 3, pp. 432–448, 2000

2000
[40]

Word production and comprehension in alzheimer’s disease: The breakdown of semantic knowledge,

A. Martin and P. Fedio, “Word production and comprehension in alzheimer’s disease: The breakdown of semantic knowledge,” Brain and language, vol. 19, no. 1, pp. 124–141, 1983

1983

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Naturalistic Language-related Movie-Watching fMRI Task for Detecting Neurocognitive Decline and Disorder

Introduction The growing aging population poses a significant challenge to society. In China, older adults aged 60 and above constitute 18.70% of the total population [1]. Aging typically leads to neu- rocognitive decline, which may progress into neurocognitive disorder (NCD) that is more severe and goes beyond the normal cognitive aging trajectory. Globa...

work page Pith review arXiv 2025

[2] [2]

Sweet Home

Data Collection Participants. 97 Chinese older adults aged above 60 years (see Table 1 for more demographic information) residing in Hong Kong were recruited. All participants were right-handed, had normal or correct-normal visual acuity, could read Chinese, without dementia, and were eligible for fMRI scanning. Eth- ical approval was obtained from the me...

work page

[3] [3]

Approach 3.1. Language-related Statistical T-maps Extraction For each participant, we used general linear modeling to analyze their brain activation related to speech processing: Y = β0 + β1(Speech event∗ HRF ) +β2(Silence event∗ HRF ) +Headmotion Regressors+ Residual (1) where the fMRI signals Y (a 1-D time series) within each brain voxel of each partici...

work page

[4] [4]

Results and Discussions 4.1. T-maps’ Coverage of Language-related Brain Areas To demonstrate that the Speech vs Silence T-maps extracted from the movie-watching task (see Figure 3a) captured the ac- tivity of brain regions associated with language processing, we compared the distribution of voxels in the masked Speech vs Silence T-map against a validated ...

work page

[5] [5]

the Speech vs Silence T-map captured brain activity associ- ated with language processing. 4.2. Results of Cognitive Status Classification The average AUC of 500 stratified shuffle-split iterations is shown in Table 2. As expected, the performance of models based on demographic features was not bad, due to the influence of age and education on cognitive a...

work page

[6] [6]

Conclusions In this paper, we introduced and demonstrated the effectiveness of a naturalistic language-related movie-watching fMRI task for detecting cognitive decline and early NCD. The language- related fMRI features (statistical T-maps) extracted from the task, combined with demographics, achieved an averaged AUC of 0.86 for classifying participants’ c...

work page

[7] [7]

T45-407/19N)

Acknowledgements This research is partially supported by the HKSARG Re- search Grants Council’s Theme-based Research Grant Scheme (Project No. T45-407/19N)

work page

[8] [8]

Bulletin of the seventh national census (no. 5),

N. B. of Statistics of China, “Bulletin of the seventh national census (no. 5),” 2023. [Online]. Available: https: //www.stats.gov.cn/sj/zxfb/202302/t20230203 1901085.html

work page 2023