AD diagnosis requires sequential evidence acquisition under heterogeneous test costs and patient burden. Fixed-modality predictors cannot jointly decide which test to acquire and when the available evidence is sufficient for diagnosis. We propose SCOPE‑AD (Sequential Cost‑Aware Ordinal‑Belief Planning with Energy‑Based Models for Diagnostic Agents) for cost‑aware classification of cognitively normal (CN), mild cognitive impairment (MCI) and Alzheimer’s disease (AD).
Key technical components:
-
A mask‑aware ordinal model that represents uncertainty along the ordered CN → MCI → AD continuum and can handle missing test values.
-
Retrospective training records provide sampled Bellman targets for an energy‑based teacher; its action distribution is distilled into a Qwen policy network.
-
At deployment the agent selects either an acquisition or a diagnosis action under availability and budget constraints, without access to unacquired values. After each acquisition the evidence and ordinal belief are updated before the next decision.
On the ADNI cohort, SCOPE‑AD achieves 77.70 % Macro‑F1 with an average acquisition cost of $50.46, outperforming the strongest baseline by 9.34 percentage points. Full‑modality evaluation raises Macro‑F1 by only 1.89 points while inflating acquisition cost by 116.7×. These results demonstrate that selective acquisition can deliver cost‑effective diagnosis.
Review