Science1 publisher2 min readPublished
Bone density read off chest CTs tracks faster cognitive decline in 715 MESA participants
Johns Hopkins ran a deep learning bone measurement over 2,086 non-contrast chest CTs, then followed the 715 people who also had brain MRI and repeat cognitive tests. The authors say bone loss does not cause dementia.
The Scientist · Science desk
What happened
- A deep learning model built in Shadpour Demehri's Johns Hopkins lab extracted thoracic vertebral volumetric bone mineral density from 2,086 non-contrast chest CT scans collected in the Multi-Ethnic Study of Atherosclerosis.
- Lower baseline spinal bone density was significantly associated with a faster rate of global cognitive decline over the follow-up period.
- Published in Radiology, the paper stops short of causation, with the investigators stating that bone loss does not cause dementia and that both appear to follow parallel metabolic aging.
Compiled by The ScientistSomething wrong?How this is made
Why it matters
- capability A bone density value can be lifted from a chest CT ordered for another reason. That puts a candidate brain-aging variable within reach of people who never had a bone scan.
- constraint With roughly 400 people supplying the longitudinal white matter measures, the regional pattern cannot be split by sex, ethnicity or age band with useful precision until a larger set of archived scans is re-read.
- decision If bone and brain are parallel outcomes of one metabolic process, an intervention trial would have to aim at that metabolism; a drug that raises bone density would not be expected to change white matter.
- precedent The finding came from re-reading images a cohort had already collected. Other CT archives with cognitive follow-up become candidates for the same secondary analysis at no new data collection cost.
The algorithm read bone in 2,086 non-contrast chest CTs collected in the Multi-Ethnic Study of Atherosclerosis [3]. Of those people, 715 also had multimodal brain MRI and longitudinal cognitive testing, and the paper analyzes that cohort [4]. The longitudinal imaging arms are smaller again: 408 participants for white matter hyperintensity progression [5] and 405 for fractional anisotropy [6]. Divide 408 by 2,086 and the tract-level findings rest on 19.6 percent of the people whose bone the model measured [18]. The analysis cohort is 34.3 percent of them [17].
Shadpour Demehri, professor of radiology at Johns Hopkins University and the study's senior author, said: "This study is the first longitudinal secondary analysis linking baseline vertebral bone mineral density to changes in white matter structure, white matter hyperintensity progression and cognition" [13]. The work was led by postdoctoral fellow Sara Momtazmanesh [15].
Both brain markers are indirect. Hyperintensities are the bright patches on MRI that indicate small vessel disease and localized ischemic injury [9]. Fractional anisotropy is a diffusion tensor measure of whether a tract keeps its structural organization [10]. The pattern the team found was regional: participants with low vBMD had more hyperintensity burden in the corpus callosum, which the study describes as a white matter bridge supporting working memory, executive function and attention [11], and a steeper decline in fractional anisotropy in the anterior limb of the internal capsule [8].
The paper names three candidate mechanisms for the association and measures none of them: insulin resistance, dyslipidemia and menopausal shifts [12]. A secondary analysis that puts baseline imaging against later brain and cognitive measures can establish that order in time [13]. Which of the three matters most stays open.
The write-up reports the cognitive association as statistically significant but not its effect size [21]. A measurement that comes free with a scan somebody already ordered still has to add something to age and sex in the same model before it changes a clinical decision, and that comparison needs the slope estimates in the full Radiology paper [2]. Osteoporosis and cognitive impairment often occur together in aging adults, and until now the skeleton and the central nervous system have largely been studied in isolation [22].
What to watch
- Replication on a clinical CT archive, where scanner protocol and slice thickness vary more than in MESA, would test whether the deep learning vBMD measurement holds outside a research cohort.
- Direct measurement of insulin resistance or lipids in the same participants, to see whether either mediates the bone-brain association the authors describe.
- Osteoporosis treatment trials that report white matter or cognitive endpoints: a direct test of the shared-driver account.