Home › Research › Digital health tools
Speaker-Wise Dementia Screening fromNaturalistic Speech with Lexical Entropy, Semantic Drift, and Multimodal Fusion
Salah R.. Research Square, 2026-10-21
Topic: Brain imaging, connectivity and assessment
Although advances have been made in dementia symptom management, early detection during stages when interventions are most effective remains a significant challenge. Most speech-based screening methods are evaluated using controlled elicitation tasks, often without examining how feature strength changes in the years preceding formal diagnosis. This study introduces a multimodal approach for dementia risk screening from naturalistic speech, integrating four traditional lexical features, two computational markers (lexical entropy from a probabilistic language model and semantic drift from sentence-level embeddings), and eGeMAPSv02 acoustic biomarkers. The framework is assessed on the DementiaNet corpus using speaker-disjoint nested group cross-validation. The optimal configuration, late fusion with stacked blending, achieves an AUC of 0.607 at the speaker level. Time-to-diagnosis stratified analysis demonstrates statistically significant discrimination in near-diagnostic windows, with text features peaking in the 5-year bin (AUC = 0.678) and acoustic features at the year of diagnosis (AUC = 0.640, p = 0.024). Exploratory analysis of more distant temporal bins (10 and 15 years) indicates that computational features may retain discriminative value where traditional features do not, although these patterns do not reach statistical significance with the current sample size. These findings support further investigation of computational language features alongside traditional and acoustic markers in larger longitudinal studies of speech-based dementia screening.
Key findings
- Although advances have been made in dementia symptom management, early detection during stages when interventions are most effective remains a significant challenge.
- Time-to-diagnosis stratified analysis demonstrates statistically significant discrimination in near-diagnostic windows, with text features peaking in the 5-year bin (AUC = 0.678) and acoustic features at the year of diagnosis (AUC = 0.640, p = 0.024).
- Exploratory analysis of more distant temporal bins (10 and 15 years) indicates that computational features may retain discriminative value where traditional features do not, although these patterns do not reach statistical significance with the current sample size.
- These findings support further investigation of computational language features alongside traditional and acoustic markers in larger longitudinal studies of speech-based dementia screening.
New to this subject? Start with our plain-English guide: Digital health →
Study details
- Study designlongitudinal study
- Effect sizeAUC = 0.678; AUC = 0.640
- Evidence strengthModerate
How this study scored 16/100
Preprint — not yet peer-reviewed. It scored 40 on the same rubric, but non-peer-reviewed sources are capped at 16, so it shows as 16. Scored zero for sample size, studied in people and randomised / blinded because the record does not state it. We have no data on times cited, citations per year, impact for its field and year, cited by studies here, still being cited, senior author's track record, institution's track record and multi-centre collaboration, so those were left out: it earned 25 of the 62 points we could judge it on, which is 40 out of 100.
| Component | Points | Why |
|---|---|---|
| Journal tier | 2 / 10 | Tier 4 — not a peer-reviewed journal (Research Square) |
| Journal's citation record | 0 / 4 | Not a journal article, so the journal's record does not count |
| Study design | 7 / 12 | A longitudinal study following the same people over time |
| Strength of evidence | 3 / 5 | Rated “moderate” in our summary |
| Peer-review validation | 4 / 10 | Preprint — not yet peer-reviewed. Hosted on a preprint server. |
| How recent | 5 / 5 | 2026 — within 1 year(s) of the newest study here |
| Sample size | 0 / 4 | Participant count not stated — scores zero (not stated in the record — scored 0) |
| Effect reported | 3 / 3 | Effect reported with a statistic (interval, p-value, ratio or similar) |
| Studied in people | 0 / 3 | Not stated whether this was studied in people — scores zero (not stated in the record — scored 0) |
| Randomised / blinded | 0 / 2 | No randomisation or blinding stated — scores zero (not stated in the record — scored 0) |
| Times cited | — / 7 | No citation count (not found in OpenAlex) — left out of this score |
| Citations per year | — / 5 | No citation count or no year — left out of this score |
| Impact for its field and year | — / 6 | No field-weighted citation impact (OpenAlex has none yet) — left out of this score |
| Cited by studies here | — / 6 | Not yet checked against this library's citation graph — left out of this score |
| Still being cited | — / 4 | No yearly citation counts — left out of this score |
| Senior author's track record | — / 4 | Senior author's record not available — left out of this score |
| Institution's track record | — / 3 | No institution listed in OpenAlex — left out of this score |
| Multi-centre collaboration | — / 3 | Institutions not listed — left out of this score |
| Indexed in PubMed | 1 / 2 | Not in PubMed, but has a DOI |
| Free to read | 0 / 2 | Behind a paywall — scores zero |
| Points | 25 / 62 | Scored on 12 of 20 parts. We have no data for the rest, so they are left out: 25 of the 62 points we could judge it on is 40 out of 100. |
| Total | 40 / 100 | Capped to 16 because this is not a peer-reviewed article. |
What each of these 20 components measures, and what this score is not →
Read the original
Free copy unknown: no free-copy check has an answer for this record yet. That does not mean it is paywalled.
Cite this study APA, Vancouver, BibTeX · download .bib or .ris
Built from the record alzhub holds for this paper, not looked up. Volume, issue and page numbers are not held in the record. Check the citation against the original before you use it.
APA 7th
Salah, R. (2026). Speaker-Wise Dementia Screening fromNaturalistic Speech with Lexical Entropy, Semantic Drift, and Multimodal Fusion. Research Square. https://doi.org/10.21203/rs.3.rs-9795204/v1
Vancouver
Salah R. Speaker-Wise Dementia Screening fromNaturalistic Speech with Lexical Entropy, Semantic Drift, and Multimodal Fusion. Research Square. 2026. doi:10.21203/rs.3.rs-9795204/v1.
BibTeX
@article{salah2026-0e313e69,
author = {Salah, R.},
title = {{Speaker-Wise Dementia Screening fromNaturalistic Speech with Lexical Entropy, Semantic Drift, and Multimodal Fusion}},
journal = {Research Square},
year = {2026},
doi = {10.21203/rs.3.rs-9795204/v1},
url = {https://doi.org/10.21203/rs.3.rs-9795204/v1}
}
Related studies in Digital health tools
Also tagged “convolutional”
205 studies in the library carry the topic tag “convolutional”; these are this study's neighbours in a fixed, alphabetical-by-hash tour of them, not the closest in content. All 205 tagged studies.