Centered cortical thickness ratios did not show a clear gain in linear cohort prediction for 145 AOMIC test participants

This study tested whether centered left-right cortical thickness ratios added linear prediction of the AOMIC PIOP1 versus PIOP2 cohort. It used public OpenNeuro version 2.0.0 tables for 433 eligible participants. Models were fitted on 288 participants and tested on 145 other participants. The reference used age, recorded sex, estimated head size, mean thickness, and the overall hemisphere ratio. The test loss gain was 0.0550 nats per person. The approximate 95% interval was -0.0766 to 0.1346 after model fitting and penalty selection were repeated in 2000 bootstrap draws. The gain remains uncertain. The interval assumes independent participants and does not cover new cohorts. A regional thickness control was added after the first test scores. It gave a larger gain of 0.4963 nats per person. Its interval keeps the model fits fixed. A scale proof states which positive factors cancel in the centered ratios. The package preserves source bytes and hashes, plans, new code, derived features, model parameters, predictions, and draws. Independent checks passed. A full offline repeat produced 31 identical core result files. The study does not separate scanner, recruitment, processing, or biological causes.

cortical thickness · hemispheric asymmetry · cohort prediction · AOMIC · logistic ridge regression · measurement scale