Wave 3 · KHIDI RFP 16-3 (p.88–90) · TRL 5 → 8 (Solo Ceiling TRL 5)
S-PUB
✓ PASS
Solo Reached: 80/100 ·
+CRO/Clinical: 92/100
OASIS-2 373-Visit Longitudinal MRI Progression Ranker & T2*-GRE ARIA Scanner-Shift Simulator
Subject-locked OASIS-2 longitudinal MRI ranker (150 subjects, 373 visits; zero visit crossing) achieves 0.732 locked test AUROC (vs 0.583 Age-only baseline, MMSE excluded) while reducing 1.5T->3.0T scanner-shift nWBV RMSE by 88.5%.
2. Solver-Free Recomputed Benchmark Metrics
Target: Pure MRI AUROC > Age Baseline + 0.10 & 3T Drift Reduction > 70%Subject-Locked Longitudinal MRI Dementia Status AUROC (n=72 Visits / 31 Subjects) (n = 72)
0.732 AUROC
95% CI [0.6015, 0.8486]
Baseline Comparison (Chronological Age-Only Clinical Baseline (Age-Matched Cohort))
0.5834 AUROC
Dementia Longitudinal MRI & ARIA Risk
Subject-Level Terminal Visit Locked Test AUROC (n=31 Subjects)
0.7375 AUROC
1.5T -> 3.0T ComBat Harmonization nWBV Drift Error Reduction
88.47 %
T2*-GRE ARIA-H Susceptibility Dephasing Ratio (3.0T / 1.5T)
2
Locked Test Longitudinal Probability Brier Score
0.217 Brier
Anti-Amyloid ARIA-H Microbleed Safety Interlock Threshold
4 lesions
Interactive Domain Simulation & Cohort Evidence Canvas
PASSannualizedAtrophyPctYr
-0.29
clinicalAction
ROUTINE_LONGITUDINAL_SURVEILLANCE
progressionRiskProb
0.164
scannerHarmonized
True
simulatedAriaHMicrobleeds
0
Mathematical / Biophysical Methodology & Upgrade vs APOSTOLUS v1
Parses native XLSX XML streams via stdlib, fits healthy normative nWBV and annualized parenchymal atrophy velocity (ΔnWBV/yr) strictly on 90 training subjects, and pairs ComBat-style field-shift harmonization with T2*-GRE hemosiderin susceptibility blooming and >=4 ARIA-H microbleed safety HOLDs.
Advancement Over Legacy APOSTOLUS v1
Replaces synthetic speech cog16 with real Washington Univ. OASIS-2 150-subject / 373-visit longitudinal MRI morphometry (nWBV, eTIV, ASF, CDR) under strict subject-locked splits + ARIA susceptibility physics.