How to Read a Radiology Report You Were Never Meant to Read

Patients get their radiology reports now. In most health systems the report lands in the portal, often before anyone has explained it, sometimes before the ordering doctor has read it. That is a good change overall and it has an obvious side effect: a lot of people are reading a document that was never written for them.

Radiology reports are written by one specialist for another. The register is deliberately cautious, the vocabulary is technical, and the phrasing that sounds alarming to a lay reader is frequently the most routine thing in the document. This piece is an attempt to make the genre legible — what the sections are for, what the recurring words mean, and which parts actually carry the message.

I should be clear about where I am standing. I build machine learning systems for medical imaging; I am not a radiologist and this is not medical advice. Nothing below substitutes for the conversation with your doctor. What it should do is make that conversation better, because knowing which question to ask is most of the difficulty.

The structure, and which part matters

Almost every report follows the same skeleton.

If you read one section, read the impression. The findings section is an inventory — it describes what is there, normal and abnormal, in anatomical order, and it is written so another clinician can verify the reasoning. Reading it cold is how people end up frightened by a sentence describing a completely normal liver.

The impression is where the radiologist says what they think it means. It is usually numbered, ordered by importance, and it contains any recommendation.

The comparison line deserves more attention than it gets. In imaging, change over time is often more informative than any single appearance. A report that says "unchanged from the prior study of March 2024" is doing significant work. A report that says "no priors available for comparison" is telling you the radiologist was working with one data point, which is why a follow-up scan is so often recommended for something that turns out to be nothing.

The vocabulary

Most of the confusion comes from a small set of words that sound worse than they are.

Two phrases cause disproportionate alarm and should be read carefully.

"Cannot be excluded" is the language of formal completeness, not of suspicion. It means the imaging does not definitively rule a thing out. A report may say malignancy cannot be excluded for a finding the radiologist considers very probably benign, because the scan cannot prove a negative.

"Nonspecific" means the appearance is compatible with several causes. It is a statement about the limits of the image, not a hedge about something ominous.

Reading a report: worked example

INDICATION: Cough and fever, 8 days.

TECHNIQUE: Frontal and lateral chest radiographs.

COMPARISON: None available.

FINDINGS: Focal consolidation in the right upper lobe with air
bronchograms. Remaining lungs are clear. Heart size is normal.
No pleural effusion. No pneumothorax. Visualised osseous
structures are unremarkable.

IMPRESSION:
1. Right upper lobe consolidation, consistent with pneumonia
   in the appropriate clinical setting.
2. Follow-up radiograph recommended after treatment to confirm
   resolution.

Most of the findings section is a list of things that are normal — the other lung, the heart, the pleural spaces, the bones. That is deliberate. Documenting the absence of a pneumothorax is part of answering the question properly.

The impression contains the qualifier "in the appropriate clinical setting," which is the radiologist saying the image is consistent with pneumonia but the diagnosis belongs to the doctor who has examined the patient. And the recommendation is not a worry signal; confirming that a consolidation clears is standard practice, partly because a consolidation that does not clear needs a different explanation.

Incidental findings, which are now very common

Modern cross-sectional imaging is sensitive enough that scanning almost any body region turns up something unrelated to the reason for scanning. Thyroid nodules, renal cysts, adrenal nodules, hepatic lesions, small pulmonary nodules — these are found constantly, and the overwhelming majority are of no consequence.

This is genuinely double-edged. Some incidental findings matter and catching them early is a real benefit. But the cascade of follow-up imaging, biopsies and anxiety that others set off is a recognised harm with a literature of its own, which is why the specialist societies have spent years writing guidelines about when not to chase something.

Pulmonary nodules are the best-developed example. The Fleischner Society's 2017 guidelines for incidental pulmonary nodules found on CT raised the size thresholds for routine follow-up compared with earlier versions, and deliberately expressed follow-up intervals as ranges rather than fixed dates so that clinicians could weigh individual risk. Broadly, very small solid nodules in low-risk patients do not require routine follow-up at all; larger ones get an interval CT; larger still prompts more active assessment. Risk factors — smoking history, prior cancer, nodule appearance and location — shift the recommendation as much as size does, which is exactly why a size number on its own does not tell you what to do.

Thyroid nodules are similar in character. They are found extremely often, and the American Thyroid Association's management guidelines put the proportion that prove malignant at roughly 7 to 15 percent, varying with age, sex, radiation history and family history. Most do not need biopsy; ultrasound characteristics and size determine which do.

The general shape of the answer is the same across organ systems: size alone is a poor guide, the appearance and your individual risk profile matter more, and there is usually a published framework your radiologist is following.

Where AI has entered the picture

This part is my own field, so a plain description.

A growing number of departments run algorithms alongside their readers. In practice these do one of three things: reorder the worklist so a study with a suspected critical finding is read sooner, mark a region for the radiologist's attention, or compute a measurement automatically — nodule volume, cardiothoracic ratio, lesion size compared with a prior.

Two things are worth knowing as a patient. First, in essentially all current deployments the radiologist signs the report and owns the interpretation; the algorithm is an instrument, and one whose flags get overruled routinely. Second, performance figures quoted for these systems come from validation studies, and the performance at your hospital on your scanner is not guaranteed to match — which is why serious departments monitor them locally rather than trusting the brochure.

If a report mentions AI-assisted analysis, it is reasonable to ask whether the finding was confirmed by the radiologist. The answer should be yes.

Questions worth asking

Not a script, just the ones that reliably get you the information the report does not spell out.

Does this finding explain my symptoms, or is it separate from them?

Does it change anything about my treatment?

If follow-up is recommended, what exactly, when, and what would make it urgent sooner?

Were there prior images to compare against, and if not, does that affect how confident we can be?

What should I watch for in the meantime?

And the one people skip: is there anything in this report that does not need to worry me? Radiologists write down a lot of things because their job is to write down everything. Getting a clinician to separate the signal from the inventory is the single most useful thing you can do with fifteen minutes of their time.