Shile Qi's group in the artificial intelligence department of Nanjing University of Aeronautics and Astronautics, with Vince D. Calhoun and other US researchers, trained the model on scans of healthy people. Most came from UK Biobank, 39,679 people, the rest from the Human Connectome Project and the Brain Genomics Superstruct Project. After age correction the mean error ranged from just over one year to under four, depending on the sample.
The model was then applied to 2,698 patients, each paired with a healthy control of the same age. The nine disorders come from different archives. Alzheimer's and mild cognitive impairment come from the US ADNI database, schizophrenia and bipolar disorder from the BSNIP-1 cohort, major depression from four Chinese hospitals. Alcohol and tobacco dependence come from a cohort in Albuquerque, ADHD and autism from two public archives of children and adolescents, ADHD-200 and ABIDE II.

The paper reports results as Cohen's d, which measures the distance between two groups in standard deviations. A d of 0.2 is small, one close to 1 is large. In Alzheimer's, 361 patients, d is 0.97, with a 95% confidence interval of 0.82 to 1.13. These patients' brains look older by almost a full standard deviation. In alcohol and tobacco dependence d lies between 0.6 and 0.7, rising to 0.84 when the two combine. Schizophrenia and bipolar disorder sit around 0.5, major depression at 0.28.
The paper does not convert these gaps into years. The headline on the press release, re-run by ScienceDaily on 29 September, speaks of faster brain ageing. Yet the study looks at each person only once. It measures how old the brain looks at one moment, not how fast it is ageing. The authors list this among the limitations.
In ADHD and autism the gap does not appear. d is 0.01 for ADHD, in 344 young people, and 0.06 for autism, in 484. The intervals stay tight around zero. They rule out large differences, not small ones or ones confined to subgroups. Split by age band, the two developmental conditions still show no difference from controls.

Each group of disorders leaves its mark in different places. In psychiatric disorders a fronto-temporal network weighs most. In addiction the default mode network, the salience network, the putamen and the thalamus matter. In dementia a fronto-occipital network does. The prefrontal cortex appears in all of them. To link these maps to genes the team used the Allen Institute's gene-expression atlas, drawn from six brains donated after death. The authors treat these regions as exploratory.

The most serious limitations are in the paper itself. Addiction and psychiatric disorders often occur together, and the analysis did not account for the overlap. Ethnicity, chronic illness, lifestyle and environment were not available across all the databases. Each disorder comes from different scanners and countries. Matched controls reduce that problem without removing it.
The work was funded by Jiangsu province's research plan and the National Natural Science Foundation of China. The authors declare no competing interests. They also state that no artificial intelligence tools or technologies were used in the study or in writing the article. The sentence jars with the method. Brain age is estimated with XGBoost, a machine-learning algorithm, and the regions are interpreted with another tool from the same family. The statement most likely means generative AI, but it does not say so.
The scan measures a shape. It shows where the brains of people with these conditions depart from the average. It does not yet say why, or in what order things happen.

Gap between MRI-estimated brain age and real age in patients versus matched controls, for nine disorders and for combined alcohol and tobacco use. The measure is Cohen's d with 95% confidence interval; grey marks intervals that include zero. Number of patients in brackets. In-house chart. Chart produced by ScienceOnline from data in Liang C. et al., PLOS Medicine 2026, doi 10.1371/journal.pmed.1004860.

References
Liang C., Pearlson G., Bustillo J. et al., “Brain aging patterns among nine neurological disorders: A case-control study”, PLOS Medicine, 23, 7, e1004860, 21 July 2026, open access, CC BY licence. doi: 10.1371/journal.pmed.1004860
Code and minimal data: Zenodo. doi: 10.5281/zenodo.20743298 and 10.5281/zenodo.20743260
PLOS, press release, re-run by ScienceDaily as “Brain Scans Reveal Which Disorders Are Linked to Faster Brain Aging”, 29 September 2026 (used only for comparison with the paper).