What Are MRI Scales and What Do They Measure?

MRI scales are standardized scoring systems that radiologists and clinicians use to rate what they see on magnetic resonance imaging scans, turning subjective visual impressions into structured grades or numbers. Rather than one universal scale, dozens of different MRI rating systems exist, each tailored to a specific organ, disease, or tissue feature. A neurologist checking for early Alzheimer’s disease uses a different scale than a urologist evaluating a suspicious prostate lesion, and both differ from what a cardiologist applies when investigating inflammation of the heart muscle. What ties them together is a shared goal: giving doctors a common language to describe findings, track changes over time, and make decisions about treatment.

How Visual Rating Scales Work in the Brain

Some of the oldest and most widely used MRI scales focus on the brain, specifically on small-vessel disease and the bright spots that appear on certain MRI sequences in the brain’s white matter. These bright spots, called white matter hyperintensities, are common in older adults and linked to stroke risk, cognitive decline, and dementia. The Fazekas scale is the standard tool for grading them, and it remains one of the most frequently cited MRI scales in clinical research.¹ The scale works simply: a radiologist looks at the scan, assesses how extensive the white matter changes are, and assigns a score from 0 (none) to 3 (large confluent areas). One study of over 300 community-dwelling individuals used both the Fazekas scale and precise volumetric measurements to evaluate how well the simple visual grade captured the actual burden of disease.²

Not every visual scale grades by total burden, though. An alternative approach rates white matter changes based solely on the size of the single largest lesion, regardless of where it sits in the brain. This method collapses the distinction between lesions near the brain’s ventricles and those deeper in the white matter, producing one global score from axial images.³ The appeal is speed: a busy clinician can glance at a scan and assign a grade in seconds. The trade-off is that a single large lesion and many scattered smaller ones could receive the same score despite representing different disease patterns.

Scales for Diagnosing Dementia

When dementia is suspected, radiologists look beyond white matter changes to measure how much specific brain regions have shrunk. Different types of dementia attack different parts of the brain, so MRI scales have been developed to rate atrophy in regions that correspond to particular diseases.

The medial temporal atrophy (MTA) scale focuses on the hippocampus and surrounding structures, the regions most affected in Alzheimer’s disease. A study comparing patients with probable Alzheimer’s to age-matched healthy volunteers found that patients had significantly greater medial temporal lobe atrophy on visual assessment, and that the degree of atrophy correlated with memory performance.⁴ The MTA scale runs from 0 (no atrophy) to 4 (severe atrophy with widening of the surrounding fluid spaces), and it tends to pick up changes relatively early in the disease, even before generalized brain shrinkage becomes obvious.

Alzheimer’s also affects the back of the brain, particularly the parietal lobes. A separate posterior atrophy (PA) scale was developed to capture this pattern. In validation work, patients with Alzheimer’s scored roughly twice as high on the PA scale as healthy controls, and the scale also distinguished Alzheimer’s from other types of dementia.⁵ The Koedam scale is a closely related tool specifically designed to evaluate parietal lobe structural changes in Alzheimer’s.⁶

These scales are most powerful when used together. Medial temporal atrophy and parietal atrophy are characteristic of Alzheimer’s, while asymmetric frontal lobe shrinkage and temporal pole atrophy point toward frontotemporal dementia. White matter hyperintensity grading, meanwhile, helps identify vascular contributions to cognitive decline.⁷ In clinical practice, a radiologist assessing a patient with memory complaints will often report scores on multiple scales simultaneously, building a profile of atrophy patterns rather than relying on any single number.

MRI Criteria in Multiple Sclerosis

Multiple sclerosis (MS) has a different relationship with MRI scales than most conditions. Instead of rating severity on a gradient, MRI criteria in MS are fundamentally diagnostic: they help determine whether a patient has the disease in the first place. MRI-based criteria were first formally incorporated into the MS diagnostic pathway in 2001 and have been revised several times since.⁸

The central concept is demonstrating that lesions are “disseminated in space and time,” meaning they appear in multiple characteristic locations in the central nervous system and that new lesions have formed at different time points. The criteria specify which brain and spinal cord regions count, how many lesions must be present, and how follow-up scans can confirm disease activity. Spinal cord imaging has become increasingly important in these guidelines, since some patients show their earliest MS changes there rather than in the brain. The result is less of a numeric grade and more of a structured checklist, but it serves the same purpose as other MRI scales: standardizing how doctors interpret images and reducing diagnostic ambiguity.

Cancer Reporting Systems

Oncology has developed some of the most structured MRI rating frameworks, organized around specific organs and designed to communicate cancer risk in clear, actionable categories.

Prostate Imaging

The Prostate Imaging Reporting and Data System (PI-RADS) assigns a score from 1 to 5 to suspicious findings on multiparametric prostate MRI, with 1 meaning clinically significant cancer is highly unlikely and 5 meaning it is highly likely. The system was developed through expert consensus to standardize how prostate MRI is acquired, read, and reported, and its second version (PI-RADS v2) represents the most current guidance on these practices.⁹ The score directly influences whether a biopsy is recommended: a PI-RADS 1 or 2 typically means monitoring, while a 4 or 5 usually triggers a biopsy. A score of 3 falls in a gray zone where clinical judgment and additional factors determine the next step.

Liver Imaging

For the liver, the Liver Imaging Reporting and Data System (LI-RADS) categorizes observations in patients at risk for liver cancer. It relies on a set of “major features” visible on contrast-enhanced MRI, including patterns of how a lesion enhances with contrast dye, whether it has a surrounding capsule, and whether it has grown over time. Individual features have varying accuracy: arterial phase enhancement, for example, was found to be highly sensitive for hepatocellular carcinoma (picking up about 89% of cancers) but not very specific on its own, while the presence of a capsule was far more specific (about 99%) but caught only about a third of cases.¹⁰ Combining multiple features is what gives the system its diagnostic power, and the final LI-RADS category (ranging from LR-1, definitely benign, to LR-5, definitely cancer) integrates all of them.

Breast Imaging

The Breast Imaging Reporting and Data System (BI-RADS) applies a similar categorical framework to breast MRI, rating findings from 0 (needs additional evaluation) through 6 (known cancer). The system standardizes how radiologists describe mass shape, margins, and enhancement patterns. Research has confirmed that specific features carry more diagnostic weight than others: irregular shape, non-circumscribed margins, rim or heterogeneous enhancement, and a washout pattern on delayed imaging were all independently associated with malignancy in a multivariate analysis.¹¹ The shared architecture across PI-RADS, LI-RADS, and BI-RADS is no accident: the “-RADS” family was designed so that clinicians across specialties would encounter a familiar reporting structure regardless of which organ was being scanned.

Musculoskeletal and Spine Scales

MRI scales for the joints and spine address a different challenge: grading degenerative changes that are extremely common in the general population and that often appear on scans of people who have no symptoms at all.

The Pfirrmann grading system rates intervertebral disc degeneration on a scale of I to V. It evaluates disc structure, the distinction between the disc’s inner and outer layers, signal brightness on MRI, and disc height.¹² Grade I is a healthy, well-hydrated disc; Grade V is a collapsed, desiccated one. The system is widely used in spine research and surgical planning, though a Grade III or IV disc in a person with no back pain is a reminder that what shows up on MRI does not always correspond to what the patient feels.

For cervical spine narrowing, the Kang grading system evaluates how much the spinal canal is compressed based on sagittal MRI images. It has been validated as providing objective and reproducible assessments, with good agreement between radiologists. Clinicians outside radiology showed slightly lower reproducibility but still found it reliable enough for clinical decisions.¹³

Knee osteoarthritis has its own ecosystem of MRI scoring. The Whole-Organ MRI Score (WORMS) was designed to evaluate multiple features of the knee simultaneously, including cartilage, bone marrow lesions, meniscal tears, and ligament integrity, rather than looking at any single structure in isolation.¹⁴ A refined successor, the MRI Osteoarthritis Knee Score (MOAKS), expanded on WORMS by improving how bone marrow lesions and cartilage are delineated and by adding features like meniscal hypertrophy and partial maceration to the scoring.¹⁵ These whole-joint scores are primarily used in research settings and large clinical trials rather than in everyday clinical reports, because they take considerable time to complete.

Cardiac MRI Criteria

Heart imaging has developed its own diagnostic framework for myocarditis, the inflammation of heart muscle that can follow viral infections or other triggers. The Lake Louise Criteria are the established standard for diagnosing myocardial inflammation on cardiac MRI. The original criteria relied on detecting tissue swelling and scarring. The updated version added quantitative mapping techniques, which measure how quickly the MRI signal decays in heart tissue (known as T1 and T2 relaxation times). Abnormally elevated T1 or T2 values suggest inflammation, and the revised criteria propose that strong evidence for myocarditis requires at least one marker of tissue swelling (based on T2 imaging) combined with at least one marker of tissue injury (based on T1 imaging, extracellular volume, or late gadolinium enhancement).¹⁶

Late gadolinium enhancement, the last element in that list, deserves a brief explanation because it appears in multiple cardiac MRI applications. After a contrast agent containing gadolinium is injected, healthy heart muscle clears it quickly, but scarred or inflamed tissue retains it. On delayed images taken about ten minutes later, areas of retained gadolinium appear bright. The pattern and location of this brightness help distinguish myocarditis from a heart attack (which causes scarring in a territory fed by a specific coronary artery) and from other conditions. T1 and T2 mapping add a quantitative layer: rather than just saying “bright” or “not bright,” the radiologist gets actual numbers. In one study establishing reference values, the average healthy myocardial T1 time was about 1005 milliseconds and the average T2 time about 67 milliseconds on a specific scanner type.¹⁷ Values that deviate significantly from these reference ranges raise suspicion for disease.

Quantitative MRI Measures Beyond Visual Scales

Not all MRI “scales” are visual rating systems. A growing number are physics-based measurements that quantify tissue properties directly, removing the subjective element of a human reader’s judgment altogether.

Apparent diffusion coefficient (ADC) mapping measures how freely water molecules move through tissue. Tightly packed cells, as in a tumor, restrict water movement and produce a low ADC value. In stroke imaging, low ADC values can indicate which tissue is acutely injured and approximately how old the damage is. One study found that a low ADC value identified a brain lesion as less than ten days old with about 88% sensitivity and 90% specificity.¹⁸ ADC is now a routine part of stroke and cancer MRI protocols.

Quantitative susceptibility mapping (QSM) takes a different physical property — the magnetic susceptibility of tissue — and turns it into a map. In the brain, iron deposits are a dominant source of susceptibility in deep gray matter structures, and iron accumulation is implicated in several neurodegenerative diseases. QSM has been validated against direct tissue iron staining, confirming a strong linear relationship between the MRI-derived susceptibility values and actual iron content.¹⁹ This means QSM can serve as a noninvasive proxy for brain iron levels, a measurement that would otherwise require a biopsy.

These quantitative techniques complement visual rating scales rather than replacing them. A radiologist might use a Fazekas grade to quickly communicate white matter disease burden in a clinical report while researchers in the same institution use volumetric measurements and ADC maps to study the same patients with greater precision.

Neonatal Brain Injury Scoring

Babies who experience oxygen deprivation around birth present a unique challenge. Their brains are still developing, injury patterns differ from adult stroke, and the stakes of early prediction are high because decisions about cooling therapy and rehabilitation depend on how severe the damage is. The Barkovich scoring system was designed for this context. It describes two main patterns of neonatal brain injury: one involving the deep brain structures (the basal ganglia and thalamus), and another involving the “watershed” zones between major arterial territories. The basal ganglia subscore evaluates injury to the thalamus, lentiform nucleus, and nearby cortex on a scale up to 4, while the watershed subscore rates white matter and cortical involvement up to 5, producing a maximum combined score of 9.²⁰ Higher scores predict worse neurodevelopmental outcomes, helping families and clinicians plan for the level of support a child may need.

Reliability Across Readers

Any visual rating system is only useful if different readers arrive at similar scores when looking at the same images. This “inter-observer reliability” varies depending on the scale and the structure being rated. For perivascular spaces (the tiny fluid-filled channels around brain blood vessels, increasingly studied as a marker of brain health), reliability ranged from moderate to good in validation studies. Ratings of perivascular spaces in the basal ganglia were more consistent across readers than those in the centrum semiovale, a higher brain region where the spaces are harder to see clearly.²¹

The general pattern is that scales requiring simple, binary-like judgments (present or absent, big or small) tend to produce better agreement than those requiring fine-grained distinctions. Scanner quality also matters: higher-resolution images make it easier for readers to agree. And the reader’s training plays a role, though perhaps a smaller one than you might expect. In the cervical stenosis grading study mentioned earlier, clinicians with less imaging experience scored slightly less consistently than radiology specialists, but the differences were not large enough to undermine the system’s clinical usefulness.¹³

When MRI Scores Cause Harm

There is a side to MRI scales that rarely appears in radiology textbooks: the psychological impact on patients who read their reports. Spine imaging offers the clearest example, because degenerative findings are nearly universal in adults over 40 and most are clinically meaningless. A randomized trial compared two groups of patients with low back pain. One group received a standard MRI report filled with technical grading language (disc desiccation, annular tears, facet arthropathy), while the other received a “clinically focused” report that contextualized findings and avoided alarming terminology. After six weeks, the group that received the standard report had more negative perceptions of their spine, higher levels of catastrophic thinking, less pain improvement, and poorer functional outcomes.²²

The implication is striking: the words used to communicate MRI grades can directly affect recovery. A Pfirrmann Grade III disc is an objective description of a mildly degenerated disc, but if it lands in a patient’s inbox without context, it can easily read as “your spine is damaged.” Some radiology departments are now experimenting with patient-friendly report language that includes notes like “this finding is common in people your age and often unrelated to pain.” The scales themselves are not the problem; the disconnect between what they measure and what patients assume they mean is.

Automated Scoring and the Role of AI

Much of the current research on MRI scales focuses on automating them. Visual rating scales are inherently limited by human variability, reading speed, and the difficulty of spotting subtle change between scans taken months or years apart. Automated volumetric tools can measure brain structures with submillimeter precision, and recent work has begun comparing these tools directly against expert visual ratings. An exploratory study found significant associations between automated brain volumetry and expert visual assessments for structures like the hippocampus, temporal lobe, and ventricles, suggesting that software can track atrophy progression in ways that align with what experienced radiologists see.²³ The advantage of automation is not necessarily better accuracy on any single scan but rather consistency across thousands of reads and the ability to detect small changes that the human eye might miss on serial imaging.

Efforts to automate the Koedam parietal atrophy scale using machine learning clustering techniques illustrate where the field is heading.⁶ If validated at scale, automated grading could make MRI scales accessible in settings that lack specialized neuroradiologists, potentially democratizing the kind of nuanced imaging interpretation that currently depends on expert availability. The visual scales will likely persist as clinical shorthand even as quantitative tools handle the precision work behind the scenes.

1
PubMed Central. Quantitative Relationship Between White Matter Hyperintensity Volume and Fazekas Score on Brain MRI
2
PubMed Central. Severity of white matter hyperintensities: Lesion patterns, cognition, and microstructural changes
3
American Journal of Neuroradiology. Evaluation of a Practical Visual MRI Rating Scale of Brain White Matter Hyperintensities for Clinicians Based on Largest Lesion Size Regardless of Location
4
PubMed Central. Atrophy of medial temporal lobes on MRI in “probable” Alzheimer’s disease and normal ageing: diagnostic value and neuropsychological correlates
5
PubMed Central. Visual assessment of posterior atrophy development of a MRI rating scale
6
Anais Estendidos da XXXVII Conference on Graphics, Patterns and Images (SIBGRAPI Estendido 2024). Automating the Koedam Parietal Atrophy Scale for Alzheimer’s Using MRI Features and Clustering Techniques
7
Journal of Neurosciences in Rural Practice. Evaluation of MR Visual Rating Scales in Major Forms of Dementia
8
PubMed Central. MRI criteria for the diagnosis of multiple sclerosis: MAGNIMS consensus guidelines
9
PubMed Central. Standards for MRI reporting-the evolution to PI-RADS v 2.0
10
PubMed. LI-RADS for MR Imaging Diagnosis of Hepatocellular Carcinoma: Performance of Major and Ancillary Features
11
AJR Am J Roentgenol. Grading System to Categorize Breast MRI in BI-RADS 5th Edition: A Multivariate Study of Breast Mass Descriptors in Terms of Probability of Malignancy
12
PLoS ONE. MRI Assessment of Lumbar Intervertebral Disc Degeneration with Lumbar Degenerative Disease Using the Pfirrmann Grading Systems
13
PubMed. Inter-observer reliability and clinical validity of the MRI grading system for cervical central stenosis based on sagittal T2-weighted image
14
PubMed. Whole-Organ Magnetic Resonance Imaging Score (WORMS) of the knee in osteoarthritis
15
PubMed Central. Evolution of semi-quantitative whole joint assessment of knee OA: MOAKS (MRI Osteoarthritis Knee Score)
16
PubMed. Cardiovascular Magnetic Resonance in Nonischemic Myocardial Inflammation: Expert Recommendations
17
PubMed Central. Standardized Myocardial T1 and T2 Relaxation Times: Defining Age- and Comorbidity-Adjusted Reference Values for Improved CMR-Based Tissue Characterization
18
PubMed Central. Evolution of apparent diffusion coefficient, diffusion-weighted, and T2-weighted signal intensity of acute stroke
19
PubMed. Validation of quantitative susceptibility mapping with Perls’ iron staining for subcortical gray matter
20
Pediatric Research. The predictive value of MRI scores for neurodevelopmental outcome in infants with neonatal encephalopathy
21
Cerebrovascular Diseases. Cerebral Perivascular Spaces Visible on Magnetic Resonance Imaging: Development of a Qualitative Rating Scale and its Observer Reliability
22
PubMed. The catastrophization effects of an MRI report on the patient and surgeon and the benefits of ‘clinical reporting’: results from an RCT and blinded trials
23
Nature. Longitudinal automated brain volumetry versus expert visual assessment of atrophy progression on MRI: an exploratory study