Predicting MCI Status From Multimodal Language Data Using Cascaded Classifiers
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Recent work has indicated the potential utility of automated language analysis for the detection of mild cognitive impairment (MCI). Most studies combining language processing and machine learning for the prediction of MCI focus on a single language task; here, we consider a cascaded approach to combine data from multiple language tasks. A cohort of 26 MCI participants and 29 healthy controls completed three language tasks: picture description, reading silently, and reading aloud. Information from each task is captured through different modes (audio, text, eye-tracking, and comprehension questions). Features are extracted from each mode, and used to train a series of cascaded classifiers which output predictions at the level of features, modes, tasks, and finally at the overall session level. The best classification result is achieved through combining the data at the task level (AUC = 0.88, accuracy = 0.83). This outperforms a classifier trained on neuropsychological test scores (AUC = 0.75, accuracy = 0.65) as well as the "early fusion" approach to multimodal classification (AUC = 0.79, accuracy = 0.70). By combining the predictions from the multimodal language classifier and the neuropsychological classifier, this result can be further improved to AUC = 0.90 and accuracy = 0.84. In a correlation analysis, language classifier predictions are found to be moderately correlated (ρ = 0.42) with participant scores on the Rey Auditory Verbal Learning Test (RAVLT). The cascaded approach for multimodal classification improves both system performance and interpretability. This modular architecture can be easily generalized to incorporate different types of classifiers as well as other heterogeneous sources of data (imaging, metabolic, etc.).
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.002 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it