Brain MRI measurements: how to read the report

Longevitytech.fund portfolio · BrainKey

BrainKey is a Longevitytech.fund portfolio company. We are proud to support its work on making brain MRI measurements more accessible and informative for brain-health assessment. [9]

The practical question is what the scan helps you and your clinician decide. A measurement may draw attention to a region that deserves closer review, support an explanation for symptoms or help compare suitable scans over time. Its value depends on image quality, the particular software and the clinical question. [1] [2]

What the evidence supports

• Established use: structural MRI is recommended when evaluating suspected cognitive disorders. Particular volumetric products have US FDA clearance for defined image-processing tasks with professional review. [1] [2]

• Promising findings: automated measurements can carry useful information about diagnosis or future decline in selected patients. Their contribution varies with the task, patient group and software. [3] [4] [5]

• Still unproven: these studies do not establish that routine repeat “brain age” scans improve outcomes in healthy adults. Stronger evidence would show that the exact analysis changes decisions and improves meaningful outcomes in a comparable population. [1] [4] [5]

The scan shows anatomy; software measures it

Magnetic resonance imaging, or MRI, produces images of the brain. A radiologist examines them for patterns of tissue loss, changes related to blood vessels or another structural explanation for symptoms. Volumetry adds an automated measurement: software divides the image into labeled regions, a process called segmentation, and calculates their volumes. Some packages also measure certain tissue abnormalities, called lesions, if the required types of MRI images are available. [1] [2]

The hippocampus, a structure involved in memory, often appears prominently in these reports. But the pattern across the brain matters more than a single number. The Alzheimer's Association guideline explains that patterns of atrophy, or tissue loss, can support a diagnosis without being specific to one disease. An MRI lacking a typical pattern does not exclude underlying disease. [1]

Software can make a measurement easier to reproduce and flag a region for review. It cannot tell the whole story of when symptoms began, whether daily tasks have changed or what cognitive testing shows. Those details remain part of interpreting the images. [1] [2]

A percentile is a comparison, not a verdict

First identify what each report column measures. A raw volume, a volume adjusted for head size and a percentile answer different questions. The exact adjustments depend on the software. [2] [3]

As a hypothetical example, imagine a hippocampal measurement at the 10th percentile. Roughly 10% of the relevant reference distribution lies at or below that value. It does not mean that 90% of the hippocampus has been lost, that Alzheimer's disease is 90% likely or that memory works at 10% of normal. This is an illustration of the statistic, not a patient report. [2] [3]

A smaller-than-average measurement on one scan also cannot show how much tissue someone has lost. A naturally smaller structure and one that has become smaller may produce the same current value. To establish change, the earlier state and the reliability of the comparison matter.

Ask whose scans form the reference group, how the result was adjusted and whether the method has been tested in people of your age and clinical background. A percentile only makes sense relative to that particular comparison. [2] [3]

Why two programs may give different answers

Programs can draw anatomical boundaries differently and use different reference datasets. In a 2022 study, researchers applied NeuroQuant, DeepBrain and FreeSurfer to the same scans from 145 people at one center and another 130 participants in the Alzheimer's Disease Neuroimaging Initiative. Regional volumes differed, and agreement between percentiles from the two commercial programs was limited in many regions. [3]

That result concerns those programs and datasets. It does not make every automated measurement useless. It does mean that numbers from different programs should not be assumed interchangeable.

For a repeat scan, record the scanner, scanning protocol, software name and version. Ask whether earlier images were reprocessed with the same method and whether the images and segmentation were checked. NeuroQuant's FDA documentation, for example, specifies imaging protocols and requires professional review of the output. [2]

Three qualities need separate assessment: repeatability means obtaining similar measurements under similar conditions; accuracy means measuring the intended feature correctly; clinical usefulness means helping make a better decision. One does not guarantee the others.

Before treating a small change as improvement or decline, ask whether it exceeds expected measurement variation for that region and method. There is no universal percentage change that makes every report clinically important. Comparing percentiles from different systems may suggest a trend that the underlying images do not support. These method comparisons do not establish how often someone should have another scan. [2] [3]

Does measurement help a clinician make a diagnosis?

A 2025 study involved 366 selected memory-clinic patients with subjective cognitive decline, mild cognitive impairment or dementia. Subjective decline means concern about worsening cognition without the objective impairment required for the other diagnoses. Two neuroradiologists—specialists in brain imaging—assessed the scans with and without NeuroQuant hippocampal percentiles. [4]

Adding the percentiles did not significantly improve their overall ability to distinguish dementia from subjective decline. The automated measurement alone performed better on that comparison, and assistance did help in the subgroup older than 65. Separating Alzheimer's disease from other causes was much harder. The useful result in older patients applied to that subgroup; it did not overturn the overall result. [4]

A measurement can therefore contain useful information without improving every clinician's assessment in every setting. This study tested a selected clinical population and two experienced readers, not screening in otherwise healthy adults or long-term benefit from receiving a report. [4]

If symptoms and the report do not agree, the next step is to examine the discrepancy: image quality, segmentation, clinical history and whether another assessment would resolve the uncertainty. The more sophisticated-looking result should not automatically decide the case. [1] [2]

Can the measurements forecast future decline?

A 2024 study followed 156 patients with subjective cognitive decline or mild cognitive impairment for an average of 32.6 months, roughly 33 months. They came from an eligible group of 297, so follow-up was incomplete. Smaller hippocampal or whole-brain measurements were associated with worsening on a combined cognitive and functional scale. [5]

Moving into a more impaired diagnostic category was a separate outcome. The hippocampal association with that outcome was no longer statistically significant after accounting for baseline cognitive-test performance; the whole-brain association also lost significance with adjustment. A separate analysis of 120 participants that included APOE, a gene relevant to Alzheimer's risk, found a statistically significant hippocampal association. [5]

The main analyses used different complete-data groups: 141 for diagnostic-category change and 117 for progression on the combined scale. They should not be treated as one uniform prediction result. The study gives useful evidence about associations in people who already had cognitive concerns, rather than a personal countdown for a healthy adult. A forecast needs testing in comparable people, including whether predicted risks match what actually happens. [5]

“Brain age” can describe very different measurements

An age-like score is a model's output, not a direct observation of the brain's calendar age. Ask which information the model uses and what prediction was tested. Putting a volume percentile beside an age estimate does not turn it into a validated dementia forecast.

A widely discussed 2025 Nature Medicine study estimated organ ages from blood proteins in 44,498 UK Biobank participants. Its protein-based brain-age estimate was associated with later health outcomes. The researchers also examined associations with MRI findings. But the organ-age model was built from blood proteins; those analyses did not validate a commercial MRI volumetry product or an eye-measurement device. [6]

A lower score after a lifestyle change also needs interpretation. It could reflect biology, measurement variation or the model's behavior. By itself, it does not prove that dementia risk fell or that healthy brain life was extended. That would require evidence connecting an intervention's effect on the score with meaningful health outcomes. [5] [6]

The exact product and intended use matter

On August 22, 2024, the US FDA cleared NeuroQuant under K241098 for specified labeling, visualization and volumetric measurement of brain structures and lesions, including reference comparisons. It is a prescription device whose clinical output must be reviewed by a radiologist or neuroradiologist. This is a defined image-processing use, not permission for every company's brain-age claim or proof of a dementia-prevention benefit. [2]

BrainKey also develops MRI-based analysis. Its public website describes this work and says its online information is educational and not intended for diagnosis. That statement concerns website information; it does not determine the regulatory status of every product or service. [7]

An August 2026 study used BrainKey measurements in 51 active and retired elite athletes. It explored regional brain-volume patterns in relation to performance ratings, tests of sensory and movement function, and concussion history. The analysis was cross-sectional: it compared observations at one period without following change over time. It therefore could not determine whether a pattern reflected training, injury, pre-existing anatomy or another influence. [8]

Some statistical patterns were unstable when the sample was repeatedly reanalyzed. The authors said that region-specific accuracy and the applicability of reference percentiles to elite athletes had not been independently established. BrainKey employees and authors with ownership or consulting interests took part. [8]

This product-specific research addresses a limited question in a selected population. It is different from the evidence for NeuroQuant and does not establish dementia screening, a reliable personal brain-age forecast or benefit from repeated scanning in healthy adults. A study in athletes and another product's FDA clearance cannot supply those conclusions. [2] [8]

Authorization, clinical value and availability at a clinic are separate facts. US clearance does not establish European or Czech authorization. Before relying on a report, ask for the actual product and version, its intended clinical purpose and the authorization relevant to your location. [2]

How Healthy Longevity Clinic experts evaluate the evidence

For someone worried by a low hippocampal percentile, the useful first question is whether the finding fits a real change in memory or daily abilities. The hypothetical 10th-percentile result describes a rank within a reference group; it does not diagnose tissue loss or dementia. The clinical discussion should connect that rank with the radiologist's findings, cognitive assessment and the history of symptoms. [1] [2] [3]

For someone whose second report looks worse, a different question comes first: are the scans and analyses comparable? A change in software or reference data may alter the number. Reviewing the images and method can be more informative than responding to a color-coded label. Only then can the clinician consider whether a biological change is likely and whether it changes care. [2] [3]

The diagnostic study's lack of significant overall improvement and its benefit in people older than 65 illustrate why “the software helps” needs a specific patient group and task. Evidence that would strengthen the case for repeated scanning in well adults would show that the exact method leads to better decisions and meaningful health outcomes in that population. A younger-looking score alone would not answer that question. [4] [5] [6]

Make the review appointment about a decision

Bring the radiology report, volumetric report and earlier images if available. Ask the clinician to explain which findings fit your symptoms, examination and cognitive testing, which might be technical and which need follow-up. Incidental findings also need an explanation and a clear person responsible for any further assessment.

The next step might be reassurance with a reason, further testing or planned clinical follow-up. If another scan is proposed, clarify what question it will answer and what result would change care. Repeating a scan simply to seek a more favorable score does not establish a health benefit. [1] [2] [5]

Three questions for a consultation

1. Does this low percentile fit my symptoms and cognitive testing, and which reference group and head-size adjustments produced it?

2. If the report has changed, were the scanner protocol, software version and segmentation comparable, and is the difference larger than expected measurement variation?

3. What decision would further testing or a repeat scan change, and who will review incidental or uncertain findings?

Common questions

Does the 10th percentile mean I have lost 90% of that brain region?

No. It is a position in a reference distribution, not a percentage of tissue lost or a probability of Alzheimer's disease. A single measurement cannot establish how much a region has changed. [2] [3]

Can a reassuring MRI rule out a cognitive disorder?

No. The absence of a typical structural pattern does not exclude underlying disease. Persistent memory or thinking problems still need assessment in their clinical context. [1]

Should I compare percentiles from two different software packages?

Only with careful review of the methods. Different boundaries, adjustments and reference groups can produce different values from the same images. Ask whether the scans can be compared using a consistent analysis. [2] [3]

Would a younger brain-age score prove that my prevention plan worked?

No. It may be an interesting change, but it could reflect measurement or model behavior. Demonstrating less dementia or more healthy life requires evidence beyond a score change. [5] [6]