Using deep learning to predict internalizing problems from brain structure in youth
Bibliographic record
Abstract
Internalizing problems (e.g., anxiety and depression) are associated with a wide range of adverse outcomes. While some predictors of internalizing problems are known (e.g., their frequent co-occurrence with neurodevelopmental (ND) conditions), the biological markers of internalizing problems are not well understood. Here, we used deep learning, a powerful tool for identifying complex and multi-dimensional brain-behaviour relationships, to predict cross-sectional and worsening longitudinal trajectories of internalizing problems. Data were extracted from four large-scale datasets: the Adolescent Brain Cognitive Development study, the Healthy Brain Network, the Human Connectome Project Development study, and the Province of Ontario Neurodevelopmental network. We developed deep learning models that used measures of brain structure (thickness, surface area, and volume) to (a) predict clinically significant internalizing problems cross-sectionally (N = 14,523); and (b) predict subsequent worsening trajectories (using the reliable change index) of internalizing problems (N = 10,540) longitudinally. A stratified cross-validation scheme was used to tune, train, and test the models, which were evaluated using the area under the receiving operating characteristic curve (AUC). The cross-sectional model performed well across the sample, reaching an AUC of 0.80 [95% CI: 0.71, 0.88]. For the longitudinal model, while performance was sub-optimal for predicting worsening trajectories in a sample of the general population (AUC = 0.66 [0.65, 0.67]), good performance was achieved in a small, external test set of primarily ND conditions (AUC = 0.80 [0.78, 0.81]), as well as across all ND conditions (AUC = 0.73 [0.70, 0.76]). Deep learning with features of brain structure is a promising avenue for biomarkers of internalizing problems, particularly for individuals who have a higher likelihood of experiencing difficulties.
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.
How this classification was reachedexpand
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 | 0.001 |
| 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 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| 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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".