Neuroimaging in Litigation: Reading the Record the Images Came From

In neurological litigation, imaging is usually the one piece of evidence both sides accept, and neither argues about. Two competent readers of the same CT scan will agree that there is blood between the brain and its covering, or that there is not. What they will disagree about is what the finding is worth: when it arose, what produced it, and whether the study performed was capable of showing something else. The American College of Radiology separates head-injury imaging into an acute phase of zero to seven days, a subacute phase under three months, and a chronic phase beyond that, and gives CT and MRI different work in each.¹ What the study was asked, and what it was equipped to answer, is set in the imaging record: which study was ordered and why, which imaging sequences were run, what history the reader was given, and what earlier study was available for comparison. What follows concerns brain imaging in adults.

CT and MRI Answer Different Questions

The division of labor between the two scanners is real, and it is routinely stated backward. CT is the first study for suspected injury inside the skull; MRI earns its place when deficits persist and remain unexplained after CT. CT leads in speed, availability, and tolerance in an unstable patient, not in superior detection of blood. MRI is at least as sensitive as CT in the first hours, and superior after that.² What CT genuinely does poorly is injury to the brain tissue itself, the earliest signs of tissue starved of blood, and anything at the base of the skull, where the surrounding bone throws streaks across the picture.

What an MRI is worth in a given case depends almost entirely on which image sequences were run. A single MRI appointment produces several, each acquired with different settings and each showing different things, and which ones are run is a choice made at the time. One sequence detects tissue dying from a blocked artery within hours. Another counts chronic changes in the brain's connecting fibers. A third makes old blood visible, and it is the only one that reliably reveals the pinpoint deposits that follow some injuries. A study run without that third sequence never looked for them.

The sequence that detects a fresh stroke is where this stops being a technicality, because it is not infallible. Reported rates of a normal result in confirmed stroke run from about 7 to 23 percent across published series, and the miss is not random: it clusters in mild deficits, in the brainstem and posterior circulation, and in scans done in the first hours, where the literature still treats the relationship between timing and a missed lesion as an open question.³ An opinion resting on the premise that this sequence does not miss strokes rests on a premise the field abandoned years ago.

The Radiology Report Is a Document With a Specified Shape

A recurring premise in imaging-heavy matters is that the radiologist reads without clinical history by design, and that the blindness is a form of objectivity. The governing documents say otherwise. The College's parameter on communicating findings puts relevant clinical information second in the report's suggested format, immediately after demographics. It provides that the body of the report should identify anything that may compromise how reliably the study catches the condition, and that the report should answer the specific clinical question or state what prevents answering it. The closing impression should give a specific diagnosis where one is possible, and the list of other conditions that could produce the same appearance where that is the honest answer.⁴ The parameter for brain MRI puts the same obligation upstream: the order form that travels with the request should carry signs, symptoms, and known diagnoses.⁵

A report produced without history is a report whose reader was denied an input the parameters say should have been supplied. That is a weakness in the interpretation, not a virtue of it.

The same document draws the boundary of the radiologist's job. It asks the reader to answer the clinical question or say what prevents answering it. It does not ask the reader to reach causation. A report that describes findings and stops has done the narrower job it was written to do. It also contemplates litigation directly, recording that patients are sometimes referred for imaging by insurers, employers, or lawyers, and that whatever the source of the referral, the interpreting physician should make every effort to see that unexpected or serious findings reach the patient.⁴ In a litigation-referred study, it becomes a discrete question about what was communicated, and to whom.

Every one of these parameters states on its face that it is not intended to establish a legal standard of care, and cautions against its use in litigation where a practitioner's decisions are in question. The same passage adds that departing from the guidance does not by itself imply the approach fell below the standard of care, and that a practitioner who departs substantially from it may document the reason in the record, which is itself a document to look for. The College's own footnote to that disclaimer cites an appellate concurrence holding that published specialty guidelines help determine the duty owed.⁵ The disclaimer is a two-sided instrument, and both edges sit in the same document.

What the Imaging Record Has to Document

The imaging record is not one document. It includes the final report, any preliminary read issued ahead of it, and any addendum issued after. It also includes the order form, which travels separately and is not usually produced unless requested by name; the line naming what prior studies the reader had; and the images themselves, a separate production where the list of image sequences actually lives. And it is the identity of the physician who read the study. A general radiologist, a brain-imaging subspecialist, and an overnight remote reader covering from another state are three different readers, and the name signed to the final report may not be the one who first read the images.

That record has six things to establish. Their absence is itself an analytical fact.

• Modality and indication. What was ordered, at what point in the acute, subacute or chronic phase, and against what clinical question.¹ A CT obtained in the emergency department and never followed by MRI, in a patient whose deficits remain unexplained weeks later, is a record with a gap in it, and the gap has a date.
• The image sequences actually run. Whether a blood-sensitive sequence was among them determines whether the pinpoint deposits could have been seen at all. Whether the stroke-detection sequence was obtained, and at what interval, determines what a negative result is worth, against a miss rate that rises in the brainstem, in mild deficits, and in the first hours.³ The report's technique line answers this when it exists; when it does not, the list has to be read off the images, which means the study itself has to be obtained.
• A prior study, or a documented search for one. Absence of a pre-incident baseline is the most consequential feature of most imaging records, and it limits inference in both directions. It does not license attribution, and it does not defeat causation. It bounds what the comparison can support.
• The clinical history supplied to the reader. A reader who was told the patient fell, and a reader who was told nothing, produce different reports of the same images, and only one of those two facts is usually in evidence.^5,4 The order form has to be requested by name.
• The impression in full, with the body's stated limitations. Not the sentence that suits the theory, but the paragraphs the radiologist wrote.⁴
• Technical adequacy. Magnet strength is recorded on every study and sets how small a finding the study could resolve; two strengths are in routine use. A study limited by patient motion is a common and checkable finding, and the parameter's limitations clause is where a reader is expected to say so.⁴

A Finding Is Not a Date and Not a Cause

Imaging findings evolve on schedules, and the schedules are windows rather than timestamps. Blood inside the skull is bright on CT when fresh and fades over days to weeks.² The signature of a fresh stroke fades on its own timetable, passing through a phase where the numbers look normal while the tissue stays injured. A finding consistent with an event inside a stated window is usually supportable. An opinion dating the same finding to forty-eight hours is not.

Causation is the harder half, and it is where base rates do the work adjectives cannot. Bright spots in the brain's connecting fibers are the most frequently litigated finding of this class, and how often they appear in people who have had no collision at all is the starting point of any honest analysis. In a community sample of adults in their mid-forties, recruited at random and imaged for research rather than for symptoms, half already had them.⁶ The finding is not evidence of an event. It is the background against which an event has to be shown.

The modern form of the comparison places a measured volume against a reference range for age and sex, and those tables now exist.⁷ They carry limits worth knowing. They rest on software measurement, so a report reading "scattered hyperintensities" yields no number to place against them, and the comparison exists only if someone runs that measurement on the archived study. The reference populations are predominantly white and drawn from developed countries, and people who volunteer for imaging research are healthier than the general population, which pushes the range down. The consensus term is hyperintensities of presumed vascular origin, and labels naming a mechanism inside the finding are discouraged for exactly that reason: a report that calls them ischemic has put a causal conclusion inside what reads as a description.⁸

The base rate cuts both ways, and the second direction is named less often. These lesions are not an inevitable consequence of ordinary aging, and a substantial burden carries a materially raised long-term risk of stroke, dementia and death. An expert who assigns an entire burden to age has made a claim about that patient's vascular risk profile without measuring it.

A normal study carries its own pair of errors. Most people who sustain a concussion recover with no imaging sign at all, and the literature searching for one is built around the subset in whom something can be shown, which is not a finding about the subset in whom nothing can.⁹ Complicated mild injury is defined by a trauma-related finding on the day-of-injury CT, so concussion and normal scan are not the same category. A normal study does not establish that the patient was uninjured. It also does not establish that the patient was injured.

What the Evidence Will and Will Not Carry

Three calibrations matter, and each stops somewhere an opinion has to stop with it.

Advanced techniques are strong at the group level and unvalidated at the individual one, and that distinction is the whole of it. Diffusion tensor imaging, which maps the orientation of the brain's white matter tracts, separates injured groups from uninjured ones; no published threshold separates one person's scan from anyone else's, and the College rates it not appropriate for the chronic-symptoms scenario most litigation turns on.¹ Serious investigators contest that position and it is not settled.¹⁰ Two limits travel with any such opinion: the direction of the change is not fixed, and the measurement is scanner-dependent enough that two scans of the same person on different machines can differ almost as much as scans of two different people. No source supplies an error rate for the method applied to one litigant, which is the form the question takes in court.

Volumetric software fails in a less familiar way than the usual objection assumes. Normative databases do exist, so that argument is unavailable. FDA clearance establishes equivalence to a device already on the market rather than accuracy, so that one is unavailable too. What remains is narrower and harder to answer: packages disagree with one another, and with hand tracing, structure by structure, so the tool and the structure both have to be named before a number means anything. And the review most often cited for these tools meeting the admissibility standard was written largely by the vendor's own affiliates. Read the affiliation block before the conclusion.

The evidentiary record and the scientific record are separate records, and they have diverged. Courts have admitted this evidence more readily than the literature underwrites it at the individual level, and amended Rule 702, effective December 2023, tightened what a proponent must show on that point.¹¹ The qualifier that carries most of these findings is in isolation, and both sides tend to drop it.

Where the Reading Goes Wrong

Imaging opinions fail in a small number of recurring ways, and they fail from both directions.

• An abnormality read as its own cause. A study obtained after a collision shows bright spots, and the finding is assigned to the collision. Half of a community population in its mid forties already carries it.⁶
• Age read as a complete explanation. The mirror error, and the more comfortable one. It treats a lesion class carrying real long-term risk as an unremarkable feature of getting older, without measuring the burden against the range for that age and sex.⁷
• A normal study read as proof of no injury. Most mild injuries leave no sign, the stroke-detection sequence is negative in a meaningful minority of genuine strokes, and a study run without a blood-sensitive sequence never looked for the finding at all.^9,3
• A missing baseline treated as decisive. In one direction its absence licenses the attribution; in the other it is said to defeat causation outright. It does neither.
• Timing collapsed into a point. An opinion calling a finding acute on a study obtained three weeks later, or comparing two studies without accounting for the interval, has read an evolution as a snapshot.²
• One line of the report quoted as the report. The impression should carry alternatives where appropriate, and the body should state any limitation.⁴ An opinion quoting the phrase and not the paragraph has quoted a fragment of a document written as a whole. The bound on that criticism sits in the same literature: there is no agreed reporting standard for these findings, so a reader cannot be faulted against one that does not exist.
• Imaging read without the examination. A lesion on one side of the brain should produce weakness on the other, and a mismatch is a finding rather than a rounding error. The exception has a name: a mass can push the brain sideways against a stiff internal membrane and compress the fibers running down the far side, producing weakness on the same side as the lesion. It is real and it is rare.
• A group-level technique offered at the individual level. Advanced imaging put forward as proof that this plaintiff sustained this injury, without engaging the scanner dependence, the nonspecificity, or the absence of any published individual-level error rate.¹⁰

These are not separate failure modes. They compound. No prior study leaves the finding undated; an undated finding invites attribution to the most recent event in the history; the attribution is defended with a technique validated for group comparison; and the group-level technique is defended by quoting the one line of the report that fits it.

Reading an Imaging-Heavy Case in Context

The radiology profession has written down how imaging is supposed to be reviewed in litigation, and the document is not widely read outside it. The College's parameter on the physician expert witness provides that a standard-of-care opinion should rest on the information available at the time of the incident, and that imaging performed afterward should never be used to form one. It provides that serial studies should be reviewed in sequence rather than side by side with the answer already known, and that the reviewer should replicate the conditions under which the original interpretation was made. And it provides that material the reviewer requested and did not receive should be disclosed and stated for the record.¹² That hindsight rule is scoped to standard-of-care opinions. It does not bar later imaging on causation or damages, and it must not be converted into a rule that later scans cannot be considered at all.

A defensible review of a case where imaging is central, plaintiff or defense, works through a small set of specific questions.

• Which studies were obtained, on what dates, and how many days separate each from the event said to have caused the finding?
• Which image sequences were actually run, and was a blood-sensitive one among them? The technique line answers this when it exists; otherwise the images have to be obtained.
• Does the order form carry the clinical history, and does the report record what the reader was told? It has to be requested by name.
• Was a prior study obtained for comparison, and if none exists, does anything document the search for one?
• Does the opinion engage the whole impression, including any stated limitation, or only the sentence it quotes?
• Was the finding measured against a reference range for age and sex, or compared by eye? In most records it was done by eye, because a clinical report describing scattered spots produces no number, and the measurement can be run afterward on the archived study.
• If advanced imaging is offered, does the opinion state what the technique is validated to support and at what level, and does it engage what bounds it?
• Does any report or opinion state that the deficit recorded on examination corresponds to the location of the lesion, and if the two do not correspond, is the mismatch addressed or passed over?
• Was the standard-of-care opinion formed on the information available at the time, or with the later imaging already in hand?

These questions, answered in order, resolve most cases. Some that look strong on the complaint dissolve against a record in which the finding sits inside the ordinary range for the patient's age and sex, on a study read against a history the reader was actually given. Others that look strong on the defense collapse against a record in which the only study obtained was the wrong one, run at the wrong interval, without the image set that would have shown the injury claimed.

The scan answers the question it was asked. The imaging record is where that question was set, and where the limits of the answer were set with it. An opinion that reaches past the question has reached past the record.

References

Footnotes

1. Expert Panel on Neurological Imaging, Shih RY, Burns J, Ajam AA, et al. ACR Appropriateness Criteria Head Trauma: 2021 Update. J Am Coll Radiol. 2021;18(5S):S13–S36. doi:10.1016/j.jacr.2021.01.006 ↩ ↩² ↩³

2. Kidwell CS, Wintermark M. Imaging of intracranial haemorrhage. Lancet Neurol. 2008;7(3):256–267. doi:10.1016/S1474-4422(08)70041-3 ↩ ↩² ↩³

3. Nikolakakis I, Koutroulou I, Mantatzis M, Finitsis S, Grigoriadis N, Karapanayiotides T. Limitations and Blind Spots of Diffusion-Weighted Imaging in the Evaluation of Acute Brain Ischemia: A Narrative Review. J Clin Med. 2026;15(2):885. doi:10.3390/jcm15020885 ↩ ↩² ↩³

4. American College of Radiology. ACR Practice Parameter for Communication of Diagnostic Imaging Findings. Revised 2025 (Resolution 9). Reston, VA: American College of Radiology; 2025. gravitas.acr.org ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

5. American College of Radiology, American Society of Neuroradiology, Society for Pediatric Radiology. ACR-ASNR-SPR Practice Parameter for the Performance and Interpretation of Magnetic Resonance Imaging (MRI) of the Brain. Revised 2024 (Resolution 4). Reston, VA: American College of Radiology; 2024. gravitas.acr.org ↩ ↩² ↩³

6. Wen W, Sachdev PS, Li JJ, Chen X, Anstey KJ. White matter hyperintensities in the forties: their prevalence and topography in an epidemiological sample aged 44-48. Hum Brain Mapp. 2009;30(4):1155–1167. doi:10.1002/hbm.20586 ↩ ↩²

7. de Kort FAS, Vinke EJ, van der Lelij EJ, et al. Cerebral white matter hyperintensity volumes: normative age- and sex-specific values from 15 population-based cohorts comprising 14,876 individuals. Neurobiol Aging. 2025;146:38–47. doi:10.1016/j.neurobiolaging.2024.11.006 ↩ ↩²

8. Duering M, Biessels GJ, Brodtmann A, et al. Neuroimaging standards for research into small vessel disease-advances since 2013. Lancet Neurol. 2023;22(7):602–618. doi:10.1016/S1474-4422(23)00131-X ↩

9. Bigler ED. Neuroimaging biomarkers in mild traumatic brain injury (mTBI). Neuropsychol Rev. 2013;23(3):169–209. doi:10.1007/s11065-013-9237-2 ↩ ↩²

10. Douglas DB, Iv M, Douglas PK, et al. Diffusion Tensor Imaging of TBI: Potentials and Challenges. Top Magn Reson Imaging. 2015;24(5):241–251. doi:10.1097/RMR.0000000000000062 ↩ ↩²

11. Fed. R. Evid. 702; Fed. R. Evid. 101(a). In: Federal Rules of Evidence, December 1, 2023. Washington, DC: Committee on the Judiciary, US House of Representatives; 2023. uscourts.gov ↩

12. American College of Radiology. ACR Practice Parameter on the Physician Expert Witness in Radiology and Radiation Oncology. Revised 2022 (Resolution 6); Amended 2023 (Resolution 2c). Reston, VA: American College of Radiology; 2023. gravitas.acr.org ↩