Consider what a diagnosis of corticobasal syndrome actually promises. One arm stops cooperating, then stiffens; speech thins out; a hand starts behaving as though it belongs to someone else. The label describes all of that faithfully. What it does not tell you is what is happening inside the brain, and the gap is not small. At the Mayo Clinic, Shir and colleagues reviewed 113 people diagnosed with corticobasal syndrome during life and then examined at autopsy, reporting in Neurology in July 2023 that 38 percent had corticobasal degeneration, 24 percent had progressive supranuclear palsy, 15 percent had Alzheimer's disease, and the rest a scatter of other pathologies. Of the cases that met clinical criteria for possible CBD at presentation, 40 percent turned out to have it. That single statistic is why no adequately powered trial of a CBD treatment has ever been run: enrol by syndrome, and more than half your participants may not have the disease your drug is designed to attack.
A paper published on 24 January 2026 in The Journal of Prevention of Alzheimer's Disease, led by Jesus Garcia-Castro and Ignacio Illan-Gala with Adam Boxer and a large 4RTNI author group, tries to price that problem. It is a longitudinal cohort analysis, not a trial: 206 people, 106 from the 4 Repeat Tauopathy Neuroimaging Initiative (with Richardson's syndrome, corticobasal syndrome, or nonfluent variant primary progressive aphasia) and 100 with Richardson's syndrome from the old phase 2/3 davunetide trial. Everyone's baseline MRI was fed through a model built earlier by the same group, published in JAMA Network Open on 29 April 2022 from 326 participants in the UCSF Memory and Aging Center database, where the atrophy patterns had been anchored to what pathologists found after death. Participants with corticobasal syndrome who carried positive Alzheimer's biomarkers were reclassified out. Of the 4RTNI group, 49 were predicted to have PSP pathology and 43 CBD pathology. The interesting part is who ended up where: 76 percent of the MRI-PSP group had Richardson's syndrome, but 24 percent had corticobasal syndrome or nonfluent aphasia, and 66 percent of the MRI-CBD group had corticobasal syndrome. The scan does not just filter people out. It pulls people in from clinical categories that a syndrome-based protocol would have turned away at the door.
The authors then built what they call data-driven MRI signatures, combinations of regional thickness and volume measures chosen for the largest change over twelve months. PSP and CBD signatures shared midbrain and pontine atrophy but diverged in the cortex. The PSP signature tracked closely with twelve-month change on the PSP Rating Scale (beta of minus 0.59, p less than 0.001), which matters, because an imaging measure that drifts independently of how a person is doing is of no use to anyone. Pairing MRI-based selection with MRI-based outcomes cut the estimated number of participants needed to detect a 30 percent slowing over twelve months by 50 percent for MRI-PSP and by 87 percent for MRI-CBD, compared with clinical selection and clinical endpoints. The team reported feasibility was replicated in the davunetide dataset. An earlier version of this work, presented as a conference abstract published in Alzheimer's and Dementia on 24 December 2025 and based on a smaller subset of 84 participants, put rough absolute figures on it: a hypothetical PSP trial falling from 336 participants to 121, and a corticobasal trial falling from 1,301 to 160.
Hold that number, 1,301, next to the history. The davunetide trial, reported by Boxer and colleagues in The Lancet Neurology (published online 27 May 2014), randomised 313 people with PSP across 48 centres in six countries for 52 weeks, and found no difference between drug and placebo on either co-primary endpoint. That was PSP, the more recognisable and more commonly diagnosed of the two, and it took a global effort to fill. A corticobasal trial requiring four times as many participants is not a hard trial. It is a trial that does not happen. Which is the quiet argument of this paper: the barrier to testing treatments in CBD has not only been the absence of candidate drugs, it has been arithmetic.
For a family sitting with a diagnosis this week, nothing here changes what happens at the next appointment. This is modelling of hypothetical trials using existing cohorts, not evidence that any treatment works, and not a new diagnostic test that a neurologist can order. The people in these cohorts were not themselves examined at autopsy, so the MRI labels are predictions, however well the model was trained. Brain shrinkage measured on a scan is also not the same thing as walking, swallowing, or speaking, and regulators generally want to see the latter before approving anything. What the work does offer is a plausible route by which the first properly powered corticobasal trial could be designed at a size someone would actually fund, and a reason for people with atypical syndromes to ask their clinicians about research registries and imaging cohorts, since the enrolment rules of the future may not look like the diagnostic labels of the present.