Adding wave 1 amyloid increased squared error by 1.9% in one offline linear test of hippocampal change in 94 selected DLBS 1.3.0 adults

We tested whether wave 1 global amyloid SUVR reduced error for hippocampal volume change in separate people. We used public MRI and PET summary measures from the Dallas Lifespan Brain Study (DLBS), OpenNeuro ds004856, version 1.3.0. The selection included 187 adults, ages 40 to 89. We fitted fixed linear models in 93 people and tested them in 94 other people. Both models used age, sex, baseline hippocampal volume, head size, and the observed scan schedule. The added model also used one amyloid term. Test mean squared error increased from 4.6435 to 4.7333. Relative error reduction was -1.93%. The approximate 95% refitted bootstrap interval was -19.30% to +3.88%. The interval permits a small benefit. It does not prove that amyloid has no useful effect. All eight planned secondary intervals crossed zero. Twenty repeats of five-fold cross-validation also increased error with amyloid; the repeats share people and do not supply a significance test. This is an offline test. Stored PET processing can use later MRI, and the MRI change interval begins before PET. The package preserves the fixed local plan, source files, retrieval hashes, code, predictions, and bootstrap draws. Independent XML and QR checks passed. Fresh downloads and a full replay matched all 35 numerical result files exactly.

amyloid · hippocampus · MRI · PET · prediction · DLBS · offline validation · brain aging