How to Write a Radiology Impression for Complex Cases (With Examples)

Twenty findings, the next study already loading, and the impression still has to say something. On a trauma pan-scan or a fourth-round oncology follow-up, the findings practically write themselves while you scroll. The impression is where the report either becomes useful to the person who ordered it or turns into a second, shorter copy of the findings. At the end of a shift the second thing happens more often than any of us would admit, and it is the only part the referrer reliably reads.

What an impression is for

Ask a trauma surgeon or an oncologist how they read a radiology report and most will give the same answer: impression first, findings only if something needs checking. That is the design constraint. The impression is not a summary of the report. It is the answer to the clinical question, written for someone with ninety seconds and a decision to make.

That settles the findings-versus-impression argument. Findings are observations: what you saw, where, how big, compared with what. The impression is interpretation: what those observations mean together, which matters most, and what should happen next. If a sentence in your impression could move into the findings without anything changing, it was never an impression statement. A complex study does not need a complex impression; it needs a short one that has already decided what matters.

A five-step method

1. Answer the question first. Reread the indication before you write a word. If the order says “fall from height, evaluate for injury,” the first line is the injury, or its absence. If it says “restaging,” the first line is whether the disease is better, worse, or unchanged. Everything else is subordinate to that answer.
2. Rank by clinical urgency, not anatomic order. Dictation runs head to pelvis because that is how you scroll. The impression runs from the finding that changes management tonight to the one that changes nothing. A liver laceration outranks a small pneumothorax; the pneumothorax outranks an incidental nodule; the nodule outranks a renal cyst, which usually should not be there at all.
3. Synthesize related findings into one statement. A liver laceration, hemoperitoneum, and pelvic free fluid are three lines in the findings and one injury in the impression. Rib fractures, an adjacent contusion, and a small pneumothorax are one chest wall injury. Grouping is where interpretation happens, and it is the step most first drafts skip.
4. State actionable follow-up with explicit criteria where a standard exists. “Follow-up recommended” is not actionable. “Repeat chest radiograph to assess for enlargement of the pneumothorax” is. Where published guidance exists, cite it: an incidental solid pulmonary nodule gets a recommendation per current Fleischner Society guidance, so the referrer knows it has a source.
5. Leave out the noise. Chronic, stable, and clinically irrelevant findings belong in the findings section and nowhere else. Hepatic steatosis on a trauma scan, degenerative change in an older spine, a simple renal cyst: if nobody needs to act on it, it does not earn a line. The test is whether deleting it would change anyone’s next decision.

Worked example: trauma pan-scan

A synthetic blunt-trauma pan-scan with the usual mix of major injury, minor injury, and incidental findings, abridged to the items that bear on the impression.

HEAD: No acute intracranial hemorrhage, mass effect, or skull fracture.
CERVICAL SPINE: No fracture or malalignment.
CHEST: Minimally displaced fractures of the right posterolateral 7th through 10th ribs. Small right apical pneumothorax, approximately 1 cm. Patchy consolidation in the right lower lobe consistent with contusion. 6 mm solid nodule in the right upper lobe; no prior imaging available.
ABDOMEN: Laceration of the right hepatic lobe (segments 6 and 7) approximately 5 cm in depth with surrounding hematoma; no active contrast extravasation. Moderate-volume hemoperitoneum in the right paracolic gutter and pelvis. Right adrenal gland enlarged and hyperdense, consistent with hematoma. Spleen, kidneys, and pancreas normal. Mild diffuse hepatic steatosis. 2 cm simple cyst, left kidney. Small hiatal hernia.
SPINE: Acute compression fracture of the superior endplate of L1 with approximately 20% loss of anterior height; no retropulsion. Fractures of the right L2 and L3 transverse processes. Multilevel degenerative disc disease.
PELVIS: No fracture.

1. No acute intracranial hemorrhage.
2. No cervical spine fracture.
3. Right 7th through 10th rib fractures.
4. Small right pneumothorax.
5. Right lower lobe contusion.
6. 6 mm right upper lobe nodule.
7. Right hepatic laceration, segments 6 and 7.
8. Hemoperitoneum.
9. Right adrenal hematoma.
10. Hepatic steatosis. Left renal cyst. Hiatal hernia.
11. L1 compression fracture. Right L2 and L3 transverse process fractures. Degenerative disc disease.
12. No pelvic fracture.

1. Blunt hepatic injury: 5 cm laceration of segments 6 and 7 with moderate hemoperitoneum and no active extravasation. Associated right adrenal hematoma. Surgical evaluation recommended; repeat imaging if clinical status changes.
2. Right chest wall injury: fractures of ribs 7 through 10 with a small (approximately 1 cm) apical pneumothorax and adjacent lower lobe contusion. Follow-up chest radiograph recommended to assess for enlargement of the pneumothorax.
3. Acute L1 superior endplate compression fracture (approximately 20% height loss) without retropulsion, and right L2 and L3 transverse process fractures. Spine evaluation per institutional trauma protocol.
4. No acute intracranial hemorrhage, cervical spine fracture, or pelvic fracture.
5. Incidental 6 mm solid right upper lobe nodule. Follow-up per current Fleischner Society guidance once the acute injuries are addressed.

The first draft is not wrong; every line is true. But it is a table of contents, not an interpretation. The liver injury, the finding most likely to send this patient to an operating room or an angiography suite, is item seven, below the incidental nodule. Three lines describe one hepatic injury and three more describe one chest wall injury. The steatosis, cyst, and hernia carry the same visual weight as the compression fracture.

The revision does the ranking and grouping for the reader: five items instead of twelve. The most urgent injury is first, with its associated findings folded in and a concrete next step. The negatives a trauma team cares about are kept but demoted to one line. The nodule survives because it would otherwise be lost, and its recommendation is anchored to published guidance. Everything chronic leaves the impression and stays in the findings. The measurements that carry down (5 cm, 1 cm, 20%) are the ones that inform the next decision; the rest stay upstairs.

Worked example: oncology follow-up

Oncology follow-ups fail differently. The problem is rarely urgency; it is comparison. The referrer wants one thing per lesion group: better, worse, unchanged, or new. An impression that lists measurements without the direction of change has not answered the question; one that buries a new lesion under a paragraph of stable disease has answered it badly.

Lead with the overall direction of the disease, then call out anything new or discordant, then collapse what is stable into one line. Use consistent response language: interval decrease, interval increase, unchanged, new. If your practice applies formal response criteria, apply them and say so; if not, do not invent a percentage. Name the comparison study, since the baseline and the immediately prior examination can tell different stories.

1. Interval decrease in size of the dominant right lower lobe mass and of the subcarinal and right hilar nodal disease compared with the immediately prior examination, consistent with ongoing treatment response.
2. New 9 mm hypoattenuating lesion in hepatic segment 4, not present on the prior study. Too small to characterize by CT; indeterminate. Dedicated liver MRI could be considered if the result would change management.
3. Unchanged 5 mm left upper lobe nodule, stable across three prior examinations, favored benign.
4. No new osseous or adrenal lesions.

Item two is the reason this impression exists: new, discordant with the overall response, and in need of a decision. It sits second only because item one answers the ordering question. Stable disease collapses into single lines, and the negatives are explicit because in oncology a stated negative is data.

Common mistakes

• Repeating findings verbatim. The most common failure and the easiest to spot: if the two sections could be swapped without confusion, there is no interpretation in either.
• Hedging every line. “Cannot exclude” belongs in one place if it belongs anywhere. An impression where every item is qualified gives the referrer nothing to act on and hands the decision to someone with less information than you.
• Burying the critical finding. Anatomic ordering puts the head first even when the liver is the problem. The finding that changes management tonight is item one, wherever it sits in the body.
• Missing the follow-up. An incidental finding without a recommendation will be forgotten. If a standard exists, name it. If not, say what you would do and why.
• An impression longer than the findings. It happens on complex studies more than anyone expects, usually because the impression restates everything and then adds hedging. If the impression is longer, it is not finished.

Where AI drafting fits

None of the five steps is hard to understand. They are hard to do on the twelfth complex study of a shift, when the temptation is to number the findings and call it an impression. That is the gap an AI drafting tool can close: not the judgment, which stays yours, but the first pass at ranking and grouping you would do yourself with more time.

RadPro AI was built for these cases. Paste your findings and it structures them, writes an impression that synthesizes the whole case rather than restating it, and runs a QA pass for internal consistency, catching a laterality mismatch or a measurement that contradicts the impression. It works alongside any PACS or dictation system, including PowerScribe and Dragon, with nothing to install. PHI detection runs before anything is processed and highlights what it finds for you to confirm. It is not a medical device and does not interpret images. Every draft is yours to edit and sign, or discard.

For how structuring fits a dictation-first workflow, see structured reporting without changing how you dictate. For the compliance questions, start with whether AI report drafting is HIPAA-compliant.

FAQ

How long should an impression be?

As long as the clinical question requires and no longer. A single-question study usually needs one or two sentences; a trauma pan-scan or multi-system oncology follow-up, four to six numbered statements. The ceiling is the findings section: if the impression is approaching that length, it is restating rather than synthesizing, and something should be grouped or cut.

Should I number the impression?

For complex cases, yes. Numbering forces you to decide what goes first, which is most of the work, and gives the referrer and the next radiologist a way to point at a specific item. For a single-finding study a sentence is fine. Follow your practice style where one exists, but do not let a house style talk you out of ranking.

Do I repeat measurements in the impression?

Only the ones that drive a decision: the size that sets a follow-up interval, the laceration depth that informs a grade, the change in a target lesion on an oncology comparison. Those belong in the impression because the reader will act on the number. Everything else stays in the findings, documented and available without cluttering the answer.