Predicting Cognitive Decline: A Multimodal AI Approach to Dementia Screening from Speech

Arav Sharma; Ari Gebhardt; Joseph T. Colonel; Lei Chi

REVIEW 3 major objections 6 minor 32 references

Predicting Cognitive Decline: A Multimodal AI Approach to Dementia Screening from Speech

T0 review · 3 major / 6 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 A two-tiered classifier that fuses Whisper, RoBERTa, and traditional acoustic features can screen for dementia and mild cognitive impairment from short speech recordings, the paper reports.

desk verdict The headline ranks in the abstract and conclusion contradict the paper's own table, and some listed ranks are impossible in a 37-team field; the central claim is unreliable as written. read the letter →

arxiv 2502.08862 v1 pith:LMWKX5NK submitted 2025-02-13 eess.AS

Lei Chi , Arav Sharma , Ari Gebhardt , Joseph T. Colonel This is my paper · ORCID

classification eess.AS

keywords dementiascreeningspeechanalysismultimodalmachinelearningPROCESSchallengeMMSEregressiontwo-tierclassificationmildcognitiveimpairmentacousticfeatures

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 argues that dementia and mild cognitive impairment can be screened from short clinical speech recordings by combining off-the-shelf neural embeddings with traditional acoustic measurements. The authors report test-set results on the PROCESS challenge: an RMSE of 2.7666 for MMSE prediction and a Macro-F1 of 0.5774 for three-way classification, ranking seventh and eleventh out of thirty-seven teams and beating several baselines. Their central contribution is a two-tiered classifier that first separates dementia from non-dementia and then separates healthy controls from impaired cases, which they say handles the overlap between MCI and dementia better than a flat three-way model. If correct, the work supports the idea that a non-invasive, speech-only pipeline can aid early screening in settings where specialist assessment is unavailable.

What carries the argument

The load-bearing object is the two-tiered classification architecture. A first binary classifier separates dementia from non-dementia; a second binary classifier separates healthy controls from impaired cases; the two decisions are combined to produce the final healthy/MCI/dementia label. This decomposition is paired with feature fusion of deep embeddings (Whisper encoder, RoBERTa on Whisper transcripts) and hand-crafted acoustic features (eGeMAPS, prosody from DisVoice), with SMOTE used to counter class imbalance and 5-fold cross-validation for tuning. The feature fusion is what the ablation tables use to show that acoustic information complements language information, and the two-tier structure is what the paper argues improves MCI versus dementia separation.

What would settle it

Compare the two-tier model's per-class F1 for MCI against a flat three-way classifier trained on the same features. If the two-tier MCI F1 is not higher, the central claim that the two-tier setup better differentiates MCI and dementia is refuted.

Watch

Extended reading notes

Core claim

The paper's central claim is that a multimodal feature stack—Whisper encoder embeddings, RoBERTa embeddings computed over Whisper transcripts, and the eGeMAPS and prosody acoustic feature sets—fed into support vector machines for classification and support vector regression for MMSE scoring produces competitive dementia screening on the PROCESS challenge. The two-tiered classification scheme is the mechanism proposed to solve the hardest part of the task: the MCI class, whose symptoms overlap with both healthy aging and dementia. The authors report that adding eGeMAPS acoustic features lifts Macro-F1 from 0.4306 to 0.5774 and that their best regression model reaches an RMSE of 2.7666, while simpler SVM-based models generalize better than the deep CNN and XGBoost pipeline tried first.

Load-bearing premise

The whole result rests on the PROCESS challenge test set being a fair, sufficiently large, and accurately labeled sample; with only about 157 subjects, strong class imbalance, and no reported confidence intervals, the mid-pack ranks could shift on another evaluation split.

Editorial extensions

If this is right

A speech-only screening pipeline can predict continuous MMSE scores and discrete diagnostic categories from the same features, allowing severity tracking and classification in one system.
Adding hand-crafted acoustic features to deep linguistic embeddings gives the largest classification boost, indicating that prosody and voice quality carry signal that text alone misses.
On small clinical speech datasets, kernel-based models such as SVM and SVR can outperform deep neural networks, so model simplicity is a feature rather than a limitation.
The two-tiered decomposition suggests that binary decisions with clearer clinical boundaries can be composed to solve a harder three-way task.
Because the input tasks are standard fluency and picture-description exercises, the pipeline runs on inexpensive recordings and supports the paper's claim of scalable, non-invasive screening.

Reading between the lines

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

The same feature stack could be evaluated on telephone-quality or compressed audio; a graceful degradation would extend the pipeline to remote monitoring, an application the paper does not test.
Fine-tuning Whisper or RoBERTa on larger dementia speech corpora, such as the one the paper names as future work, would reveal whether the two-tier SVM's performance is a ceiling or a floor for this approach.
A calibration study comparing the two-tier model's confidence scores against clinician triage decisions would test whether the system can be used as a screening aid rather than only a ranking device.

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,

The headline of this paper does not survive contact with its own results table. The abstract and conclusion say the system ranked seventh in regression and eleventh in classification out of 37 teams, but Section V shows the best regression rank as 9 and the best classification rank as 15. Worse, Table II lists ranks of 72, 76, and 95 for baselines in a competition with only 37 teams. Those numbers cannot all be correct. This is not a nitpick: the paper's central claim is the rank, and it is internally inconsistent.

What the paper does well: it is a straightforward, honest challenge write-up. The feature combination — Whisper and RoBERTa embeddings plus eGeMAPS and prosody, fed to an SVM cascade — is a sensible recipe for multimodal speech-based dementia screening. It is evaluated on a held-out challenge test set, and the authors acknowledge the dataset is small and imbalanced. The limitations paragraph is good.

The soft spots are more than cosmetic. The two-tier cascade is meant to be the methodological contribution, yet there is no single-tier control, so we have no evidence it helps at all. There are no error bars or confidence intervals, and no code or data release, so the reported numbers cannot be checked. The citation pattern is fine; the authors themselves cite prior cascaded classifiers, so the novelty claim is modest.

The right verdict: this is a workable pipeline, but as written the empirical core is unreliable. If the authors fix the ranks, add a single-tier comparison, and report variance, it could be a legitimate workshop paper. As submitted, I would not send it to peer review; the internal contradiction makes it unsupportable as is.

For your reading group: skip it, unless you want a case study in how not to report challenge results. I would not cite it in this form. It shows some honest awareness of limitations, but the central numbers do not hold together.

Referee Report

3 major / 6 minor

Summary. This manuscript describes a multimodal speech-based screening pipeline submitted to the PROCESS 2025 challenge. The authors combine Whisper and RoBERTa embeddings with eGeMAPS/prosody acoustic features, feed them to an SVM/SVR, and propose a two-tiered classification scheme (dementia vs non-dementia, then healthy control vs impaired) to predict one of three cognitive states and regress MMSE. The reported test-set results are a Macro-F1 up to 0.5774 and an RMSE down to 2.7666, with claimed ranks of 11th and 7th among 37 teams. The paper also describes a CNN-XGBoost alternative and three final model variants with different feature sets.

Significance. If the reported numbers and ranks are reliable, the paper demonstrates a practical multimodal screening pipeline and a potentially useful hierarchical classification strategy for the difficult MCI/dementia boundary. The work uses a held-out challenge test set rather than an in-house split, which reduces overfitting concerns, and the feature ablation across three model variants is a useful empirical contribution. However, the central rank claims are internally inconsistent, the two-tiered design's benefit is not directly evidenced, and no uncertainty estimates accompany the point estimates; the significance is therefore conditional on a corrected and supported reporting.

major comments (3)

[Abstract; Table II; Section V.A; Section VI] The headline ranks are inconsistent across parts of the paper. The abstract and conclusion state that the system ranked 7th on regression and 11th on classification out of 37 teams, but Table II reports a best regression rank of 9 (Base Model) and best classification ranks of 15 and 16 ('+ Prosody' and '+ eGeMAPS'). Table II also lists ranks of 76, 57, and 95 for baselines, which cannot be positions among 37 teams, and Section V.A says '15th overall' while the conclusion says 11th. Please re-derive all rankings, define the denominator (teams vs submitted model variants) and the ranking scheme, and make the abstract, conclusion, Table II, and Section V.A state one consistent set of numbers. As written, the numerical foundation of the headline claim is not internally coherent.
[Section IV.B; Table I; Table II; Fig. 5] The two-tiered classification setup is claimed to 'better differentiate MCI and dementia,' but no controlled comparison isolates this benefit. The eGeMAPS baselines in Table II are single-tier classifiers but use different features, so a two-tier SVM with the same feature set is never compared against a one-stage three-class SVM. Fig. 5 shows confusion matrices, yet the text does not report per-class precision/recall for the MCI-versus-dementia pair. Please add an ablation using identical features in both a one-stage and a two-stage architecture and report per-class metrics; otherwise the 'better differentiate' claim is unsupported.
[Section V; Section IV.F; Section VI] All performance claims are point estimates from a single test set, with no confidence intervals, cross-validation error bars, or statistical tests, despite the authors' own acknowledgment in Section VI that the dataset is 'small and imbalanced.' Section IV.F mentions 5-fold cross-validation, but no fold-level scores are reported. Because the margin between the best models and baselines is small (e.g., Macro-F1 0.5774 vs 0.5500), the claim of 'exceeding the baseline results' is not established as stable. Please provide uncertainty measures (e.g., bootstrap intervals or per-fold results) and discuss the test-set size and class distribution when interpreting the ranks.

minor comments (6)

[Equation (8)] The combined-score formula uses RMSE_i in the second term, but i is not defined in that context and should presumably be RMSE_k; please correct the notation and define the sums over j and T consistently.
[Section IV.A] The phrase 'Whisper’s acoustic front-end feature extractor' is unusual; please specify whether the reported embeddings are the encoder output, a hidden layer, or another internal representation.
[Section III.D] The text says the method 'employs an end-to-end model that jointly processes both modalities,' but the pipeline in Section IV uses separate feature extraction followed by SVM/SVR; please rephrase to avoid 'end-to-end.'
[Sections I, II, IV] There are several language errors: 'a incremental' should be 'an incremental,' 'the DisV oice package' should be 'the DisVoice package,' and 'composed a 1D CNN' should be 'composed of a 1D CNN.'
[Section IV.F] No hyperparameters from the grid search or the SMOTE oversampling ratio are reported; please include the selected values or a table of the searched ranges.
[Section V.B] The statement that 'the model’s performance remained stable across different combinations of features' is supported by only two RMSE values; please soften it to 'similar' or add supporting statistics.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found; the reported metrics are empirical measurements on a held-out test set.

full rationale

The paper's claimed outputs (Macro-F1 0.5774, RMSE 2.7666, and the challenge ranks) are measured predictions on the PROCESS challenge test set, not quantities reconstructed from the model's own fitting procedure. The two-tier classification is an architectural choice; its reported performance is compared against baselines and against held-out test labels, so the claim that it better differentiates MCI from dementia is an empirical assertion, not a definitional consequence. Feature selection (eGeMAPS, Prosody, Whisper, RoBERTa) relies on externally published models and standard feature sets, none of which are justified by the present authors' prior work or by the target labels. No load-bearing premise is supported by a self-citation, no uniqueness theorem is imported from the authors, and no fitted parameter is renamed as a prediction. The abstract and conclusion ranks (7th in regression, 11th in classification) are inconsistent with Table II and Section V, and Table II contains rank values exceeding the 37-team field, but these are internal consistency and correctness defects, not circularity. There is no step in the derivation chain where an output equals an input by construction, so the appropriate circularity score is 0.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The paper introduces no new theoretical entities and performs no derivations. Its results rest on standard machine learning assumptions about label quality, feature informativeness, and ASR reliability, plus a number of omitted implementation choices (SVM hyperparameters, SMOTE settings) that prevent exact replication.

free parameters (2)

SVM regularization and kernel hyperparameters = not reported
Tuned via grid search on the PROCESS training set per Section IV-F. The exact values are omitted, so the model cannot be reproduced exactly and the final results depend on these unstated choices.
SMOTE oversampling ratio = not reported
SMOTE is used to address class imbalance, but the oversampling ratio and nearest-neighbor settings are not specified. This choice affects the training distribution and the resulting classifier.

assumptions (3)

domain assumption Clinical labels (HC, MCI, dementia, MMSE) in the PROCESS dataset are clinically accurate.
The evaluation treats the provided labels as ground truth. If labels are noisy or misassigned, the reported metrics do not reflect true screening performance.
domain assumption Speech-based acoustic and linguistic features carry sufficient signal to predict cognitive status.
The entire approach assumes that dementia and MCI manifest in speech in ways detectable by the chosen features. If this premise is weak, the pipeline has no information to learn from.
domain assumption Whisper transcriptions are accurate enough to preserve dementia-relevant language features.
RoBERTa embeddings are computed from Whisper transcripts. Errors in automatic speech recognition, especially on dysarthric or accented speech, could degrade or distort the linguistic features.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Predicting Cognitive Decline: A Multimodal AI Approach to Dementia Screening from Speech." pith.science (2026). https://pith.science/paper/LMWKX5NK

@misc{pith2026250208862,
  author       = {Pith},
  title        = {Pith review of: Predicting Cognitive Decline: A Multimodal AI Approach to Dementia Screening from Speech},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LMWKX5NK}},
  note         = {Machine review of arXiv:2502.08862}
}

read the original abstract

Recent progress has been made in detecting early stage dementia entirely through recordings of patient speech. Multimodal speech analysis methods were applied to the PROCESS challenge, which requires participants to use audio recordings of clinical interviews to predict patients as healthy control, mild cognitive impairment (MCI), or dementia and regress the patient's Mini-Mental State Exam (MMSE) scores. The approach implemented in this work combines acoustic features (eGeMAPS and Prosody) with embeddings from Whisper and RoBERTa models, achieving competitive results in both regression (RMSE: 2.7666) and classification (Macro-F1 score: 0.5774) tasks. Additionally, a novel two-tiered classification setup is utilized to better differentiate between MCI and dementia. Our approach achieved strong results on the test set, ranking seventh on regression and eleventh on classification out of thirty-seven teams, exceeding the baseline results.

Figures

Figures reproduced from arXiv: 2502.08862 by the authors.

**Figure 1.** Cookie theft test prompt image II. BACKGROUND A. Cognitive Assessment Tasks Three different tasks, each intended to assess a different facet of cognitive function, are incorporated [13]. In the first task, Semantic Fluency, participants have one minute to name as many animals as they can [14]. This test is essential for assessing name skills, linguistic proficiency, and semantic memory access, all of which are frequ… view at source ↗

**Figure 2.** Confusion matrix of initial model The Cookie Theft picture description task, the third evaluation, is a commonly used technique in dementia detection studies [15]. After viewing a standardized image of a kitchen scene, participants are asked to explain the details and activities they see. This extensive task simultaneously activates several cognitive domains, such as memory integration, narrative construction, an… view at source ↗

**Figure 3.** PROCESS dataset demographic metrics B. ADReSS and ADReSSo Challenges To standardize research in this field, the Alzheimer’s Dementia Recognition through Spontaneous Speech (ADReSS) challenge (INTERSPEECH 2020) introduced a balanced dataset for AD classification and MMSE score regression [26]. This dataset mitigated demographic biases and enabled systematic comparisons. The ADReSSo challenge (2021) extended this wo… view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: Two-tiered classification architecture. were also used. Conventional features were also added to these, such as the eGeMAPS and prosodic feature sets. B. Two-Tiered Classification System Previous challenges in cognitive impairment detection typically focused on the bi…

**Figure 5.** Figure 5: Confusion matrices for our three submitted models on the test set. From left to right: Base model (Whisper + RoBERTa embeddings), Model 2 (Base [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 25 canonical work pages

[1]

Essentials of the proper diagnoses of mild cognitive impairment, dementia, and major subtypes of dementia,

D. S. Knopman, B. F. Boeve, and R. C. Petersen, “Essentials of the proper diagnoses of mild cognitive impairment, dementia, and major subtypes of dementia,” in Mayo Clinic Proceedings , vol. 78, no. 10. Elsevier, 2003, pp. 1290–1308

work page 2003
[2]

The correlation of everyday cognition test scores and the progression of alzheimer’s disease: a data analytics study,

F. Thabtah, R. Spencer, and Y . Ye, “The correlation of everyday cognition test scores and the progression of alzheimer’s disease: a data analytics study,” Health information science and systems , vol. 8, pp. 1–11, 2020

work page 2020
[3]

World Alzheimer Report 2024: Global changes in attitudes to dementia,

A. D. International, “World Alzheimer Report 2024: Global changes in attitudes to dementia,” London, England, September 2024

work page 2024
[4]

World alzheimer report 2021: Journey through the diagnosis of dementia,

S. Gauthier, P. Rosa-Neto, J. A. Morais, and C. Webster, “World alzheimer report 2021: Journey through the diagnosis of dementia,” Alzheimer’s Disease International, 2021. [Online]. Available: https: //www.alzint.org/resource/world-alzheimer-report-2021/

work page 2021
[5]

Dementia prevention, intervention, and care,

G. Livingston, A. Sommerlad, V . Orgeta, S. G. Costafreda, J. Huntley, D. Ames, C. Ballard, S. Banerjee, A. Burns, J. Cohen-Mansfield, C. Cooper, N. Fox, L. N. Gitlin, R. Howard, H. C. Kales, E. B. Larson, K. Ritchie, K. Rockwood, E. L. Sampson, and N. Mukadam, “Dementia prevention, intervention, and care,” The Lancet North American Edition , vol. 390, no...

work page 2017
[6]

Suspecting dementia: canaries, chameleons and zebras,

J. C. S. Johnson, L. McWhirter, C. J. D. Hardy, S. J. Crutch, C. R. Marshall, C. J. Mummery, J. D. Rohrer, M. N. Rossor, J. M. Schott, R. S. Weil, N. C. Fox, and J. D. Warren, “Suspecting dementia: canaries, chameleons and zebras,” Practical Neurology, vol. 21, no. 4, pp. 300– 312, 2021

work page 2021
[7]

Ai-based speech analysis for predicting progression from mci to alzheimer’s disease,

Y . Huang, J. Smith, and A. Johnson, “Ai-based speech analysis for predicting progression from mci to alzheimer’s disease,” Journal of Alzheimer’s Disease, vol. 78, no. 2, pp. 456–468, 2024

work page 2024
[8]

Connected speech and language in mild cognitive impairment and alzheimer’s disease: A review of picture description tasks,

K. D. Mueller, B. Hermann, J. Mecollari, and L. S. Turkstra, “Connected speech and language in mild cognitive impairment and alzheimer’s disease: A review of picture description tasks,” Journal of Clinical and Experimental Neuropsychology, vol. 40, no. 9, pp. 917–939, 2018

2018

Show all 32 references

[9]

Identification of mild cognitive impairment from speech in swedish using deep sequential neural networks,

C. Themistocleous, M. Eckerstr ¨om, and D. Kokkinakis, “Identification of mild cognitive impairment from speech in swedish using deep sequential neural networks,” Frontiers in Neurology, vol. 11, p. 513, 2020

2020
[10]

Automatic detection of mild cognitive impairment from spontaneous speech using asr,

L. T ´oth, G. Gosztolya, V . Vincze, I. Hoffmann, G. Szatl ´oczki, E. Bir ´o, F. Zsura, M. P ´ak´aski, and J. K ´alm´an, “Automatic detection of mild cognitive impairment from spontaneous speech using asr,” Proceedings of Interspeech, pp. 2554–2558, 2018

2018
[11]

Predicting mci status from multimodal language data using cascaded classifiers,

K. C. Fraser, K. Lundholm Fors, M. Eckerstr ¨om, F. ¨Ohman, and D. Kokkinakis, “Predicting mci status from multimodal language data using cascaded classifiers,” Frontiers in Aging Neuroscience, vol. 11, p. 205, 2019

2019
[12]

Spo- ken language derived measures for detecting mild cognitive impairment,

B. Roark, M. Mitchell, J. P. Hosom, K. Hollingshead, and J. Kaye, “Spo- ken language derived measures for detecting mild cognitive impairment,” IEEE Transactions on Audio, Speech, and Language Processing, vol. 19, no. 7, pp. 2081–2090, 2011

2011
[13]

Early dementia detection using multiple spontaneous speech prompts: The process challenge,

F. Tao, B. Mirheidari, M. Pahar, S. Young, Y . Xiao, H. Elghazaly, F. Peters, C. Illingworth, D. Braun, R. O’Malley et al., “Early dementia detection using multiple spontaneous speech prompts: The process challenge,” arXiv preprint arXiv:2412.15230 , 2024

2024 arXiv
[14]

Crawford and J

J. Crawford and J. Henry, Assessment of executive deficits , 2005, pp. 233–245

2005
[15]

Goodglass, E

H. Goodglass, E. Kaplan, and S. Weintraub, BDAE: The Boston diagnos- tic aphasia examination . Lippincott Williams & Wilkins Philadelphia, PA, 2001

2001
[16]

“mini-mental state

M. F. Folstein, S. E. Folstein, and P. R. McHugh, ““mini-mental state”: a practical method for grading the cognitive state of patients for the clinician,” Journal of psychiatric research , vol. 12, no. 3, pp. 189–198, 1975

1975
[17]

The geneva minimalistic acous- tic parameter set (gemaps) for voice research and affective computing,

F. Eyben, M. W ¨ollmer, and B. Schuller, “The geneva minimalistic acous- tic parameter set (gemaps) for voice research and affective computing,” IEEE Transactions on Affective Computing , vol. 7, no. 2, pp. 190–202, Apr. 2016

2016
[18]

Modeling prosodic features with joint factor analysis for speaker verification,

N. Dehak, P. Dumouchel, and P. Kenny, “Modeling prosodic features with joint factor analysis for speaker verification,” IEEE Transactions on Audio, Speech, and Language Processing , vol. 15, no. 7, pp. 2095– 2103, 2007

2007
[19]

Towards an automatic evaluation of the dysarthria level of patients with parkinson’s disease,

J. C. V ´asquez-Correa, J. Orozco-Arroyave, T. Bocklet, and E. N ¨oth, “Towards an automatic evaluation of the dysarthria level of patients with parkinson’s disease,” Journal of communication disorders , vol. 76, pp. 21–36, 2018

2018
[20]

wav2vec 2.0: A framework for self-supervised learning of speech representations,

A. Baevski, Y . Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representations,” Advances in neural information processing systems, vol. 33, pp. 12 449– 12 460, 2020

2020
[21]

Robust speech recognition via large-scale weak supervision,

A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust speech recognition via large-scale weak supervision,” in Proceedings of the 40th International Conference on Machine Learning , 2023. [Online]. Available: https://proceedings.mlr. press/v202/radf...

2023
[22]

Roberta: A robustly optimized bert pretraining approach,

Y . Liu, M. Ott, N. Goyal, J. Du, M. Joshi, D. Chen, O. Levy, M. Lewis, L. Zettlemoyer, and V . Stoyanov, “Roberta: A robustly optimized bert pretraining approach,” arXiv e-prints , 2019. [Online]. Available: https://ui.adsabs.harvard.edu/abs/2019arXiv190711692L/abstract

2019
[23]

Linguistic features identify alzheimer’s disease in narrative speech,

K. C. Fraser, J. A. Meltzer, and F. Rudzicz, “Linguistic features identify alzheimer’s disease in narrative speech,” Journal of Alzheimer’s disease, vol. 49, no. 2, pp. 407–422, 2015

2015
[24]

Automatic prediction of linguistic decline in writings of subjects with degenerative dementia,

D. Weissenbacher, T. A. Johnson, L. Wojtulewicz, A. Dueck, D. Locke, R. Caselli, and G. Gonzalez, “Automatic prediction of linguistic decline in writings of subjects with degenerative dementia,” in Proceedings of the 2016 Conference of the North American Chapter of the Associa...

2016
[25]

Using linguistic features longitudinally to predict clinical scores for alzheimer’s disease and related dementias,

M. Yancheva, K. C. Fraser, and F. Rudzicz, “Using linguistic features longitudinally to predict clinical scores for alzheimer’s disease and related dementias,” in Proceedings of SLPAT 2015: 6th workshop on speech and language processing for assistive technologies , 2015, pp. 134–139

2015
[26]

Alzheimer’s dementia recognition through spontaneous speech,

S. Luz, F. Haider, S. de la Fuente Garcia, D. Fromm, and B. MacWhin- ney, “Alzheimer’s dementia recognition through spontaneous speech,” p. 780169, 2021

2021
[27]

Detecting cognitive decline using speech only: The adresso challenge,

S. Luz, F. Haider, S. de la Fuente, D. Fromm, and B. MacWhinney, “Detecting cognitive decline using speech only: The adresso challenge,” arXiv preprint arXiv:2104.09356 , 2021

2021 arXiv
[28]

To bert or not to bert: comparing speech and language-based approaches for alzheimer’s disease detection,

A. Balagopalan, B. Eyre, F. Rudzicz, and J. Novikova, “To bert or not to bert: comparing speech and language-based approaches for alzheimer’s disease detection,” arXiv preprint arXiv:2008.01551 , 2020

2008 arXiv
[29]

An assessment of paralinguistic acoustic features for detection of alzheimer’s dementia in spontaneous speech,

F. Haider, S. de la Fuente, and S. Luz, “An assessment of paralinguistic acoustic features for detection of alzheimer’s dementia in spontaneous speech,” IEEE Journal of Selected Topics in Signal Processing , vol. 14, no. 2, pp. 272–281, 2020

2020
[30]

Automated screening for alzheimer’s dementia through spontaneous speech

M. S. S. Syed, Z. S. Syed, M. Lech, and E. Pirogova, “Automated screening for alzheimer’s dementia through spontaneous speech.” in Interspeech, vol. 2020, 2020, pp. 2222–6

2020
[31]

Smote: synthetic minority over-sampling technique,

N. V . Chawla, K. W. Bowyer, L. O. Hall, and W. P. Kegelmeyer, “Smote: synthetic minority over-sampling technique,” Journal of artificial intel- ligence research, vol. 16, pp. 321–357, 2002

2002
[32]

Dementiabank: Theoretical rationale, protocol, and illustrative analyses,

A. M. Lanzi, A. K. Saylor, D. Fromm, H. Liu, B. MacWhinney, and M. L. Cohen, “Dementiabank: Theoretical rationale, protocol, and illustrative analyses,” American Journal of Speech-Language Pathology, vol. 32, no. 2, pp. 426–438, 2023

2023

Pith tools

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

[1] [1]

Essentials of the proper diagnoses of mild cognitive impairment, dementia, and major subtypes of dementia,

D. S. Knopman, B. F. Boeve, and R. C. Petersen, “Essentials of the proper diagnoses of mild cognitive impairment, dementia, and major subtypes of dementia,” in Mayo Clinic Proceedings , vol. 78, no. 10. Elsevier, 2003, pp. 1290–1308

work page 2003

[2] [2]

The correlation of everyday cognition test scores and the progression of alzheimer’s disease: a data analytics study,

F. Thabtah, R. Spencer, and Y . Ye, “The correlation of everyday cognition test scores and the progression of alzheimer’s disease: a data analytics study,” Health information science and systems , vol. 8, pp. 1–11, 2020

work page 2020

[3] [3]

World Alzheimer Report 2024: Global changes in attitudes to dementia,

A. D. International, “World Alzheimer Report 2024: Global changes in attitudes to dementia,” London, England, September 2024

work page 2024

[4] [4]

World alzheimer report 2021: Journey through the diagnosis of dementia,

S. Gauthier, P. Rosa-Neto, J. A. Morais, and C. Webster, “World alzheimer report 2021: Journey through the diagnosis of dementia,” Alzheimer’s Disease International, 2021. [Online]. Available: https: //www.alzint.org/resource/world-alzheimer-report-2021/

work page 2021

[5] [5]

Dementia prevention, intervention, and care,

G. Livingston, A. Sommerlad, V . Orgeta, S. G. Costafreda, J. Huntley, D. Ames, C. Ballard, S. Banerjee, A. Burns, J. Cohen-Mansfield, C. Cooper, N. Fox, L. N. Gitlin, R. Howard, H. C. Kales, E. B. Larson, K. Ritchie, K. Rockwood, E. L. Sampson, and N. Mukadam, “Dementia prevention, intervention, and care,” The Lancet North American Edition , vol. 390, no...

work page 2017

[6] [6]

Suspecting dementia: canaries, chameleons and zebras,

J. C. S. Johnson, L. McWhirter, C. J. D. Hardy, S. J. Crutch, C. R. Marshall, C. J. Mummery, J. D. Rohrer, M. N. Rossor, J. M. Schott, R. S. Weil, N. C. Fox, and J. D. Warren, “Suspecting dementia: canaries, chameleons and zebras,” Practical Neurology, vol. 21, no. 4, pp. 300– 312, 2021

work page 2021

[7] [7]

Ai-based speech analysis for predicting progression from mci to alzheimer’s disease,

Y . Huang, J. Smith, and A. Johnson, “Ai-based speech analysis for predicting progression from mci to alzheimer’s disease,” Journal of Alzheimer’s Disease, vol. 78, no. 2, pp. 456–468, 2024

work page 2024

[8] [8]

Connected speech and language in mild cognitive impairment and alzheimer’s disease: A review of picture description tasks,

K. D. Mueller, B. Hermann, J. Mecollari, and L. S. Turkstra, “Connected speech and language in mild cognitive impairment and alzheimer’s disease: A review of picture description tasks,” Journal of Clinical and Experimental Neuropsychology, vol. 40, no. 9, pp. 917–939, 2018

2018