In radiology, the report is the claim. The report elements that change the CPT and ICD-10-CM codes, the gaps that send studies to manual review, and a 10-question audit you can run.
In radiology, the report is the claim. The coder, or the software, never sees the patient and rarely sees the images. Every CPT code, modifier and diagnosis comes from the words the radiologist dictated. That makes radiology report documentation requirements a revenue and compliance issue, not a style preference. This article covers what a complete report contains, which elements change the code, which elements drive the diagnosis, and why the same gaps that slow a human coder push a study out of an AI-assisted tool's automated path.
What a complete radiology report contains
The reference standard is the ACR Practice Parameter for Communication of Diagnostic Imaging Findings, most recently revised in 2025. It recommends this structure for the final report:
- Demographics: facility, patient identifiers, referring provider (or a self-referral statement), examination name, date, and time where relevant
- Relevant clinical information: why the study was ordered
- Procedures and materials: what was performed, contrast or radiopharmaceutical used, medications, devices, and any reactions or complications
- Findings: described in standard anatomic, pathologic and radiologic terms
- Potential limitations: anything that reduces sensitivity or specificity
- Clinical issues: an answer to the clinical question, or why it couldn't be answered
- Comparison studies: when available and relevant
- Impression: a specific diagnosis when possible, a differential when appropriate, and follow-up recommendations
The parameter allows standardized templates as long as they meet these criteria. It also sets expectations for documenting nonroutine communications, such as a pneumothorax called to the referring provider: the date, time, method and recipient belong in the report.
Why coders code from the body, not the order
The order says what the referring provider asked for. The header often repeats the order. The body says what was actually done. When they disagree, the body wins, because that's the service performed and the one the radiologist signed.
Under 42 CFR 410.32(a), Medicare requires diagnostic imaging to be ordered by the physician treating the beneficiary. That creates a second check: if the radiologist performed something different from the order (adding contrast, adding views, switching modality), Medicare allows it only in limited circumstances, and the report needs to say why. A report that silently differs from the order is a compliance flag even when the code matches the body.
A typical case: the order reads "CT abdomen without contrast, flank pain." The body reads "CT abdomen and pelvis without and with IV contrast. Initial noncontrast images showed an indeterminate 2.4 cm left renal lesion; post-contrast images were obtained for characterization." The billable service follows the body, which now covers abdomen and pelvis in two phases. Whether it's payable depends on whether the change fits Medicare's exceptions and your facility's protocol, and the sentence explaining why contrast was added is what an auditor will look for.
For the same reason, treat addenda as part of the report. An addendum that adds a finding or corrects laterality can change the code after the first version was already coded.
Report elements that change the CPT code
Most radiology code selection turns on a handful of facts that must be stated, not implied. If the report doesn't say it, the coder can't assume it.
| Element | What the report must state | Example of the coding impact |
|---|---|---|
| Number of views | The views actually obtained | Chest radiographs are coded 71045 to 71048 by number of views |
| Contrast and route | Whether contrast was given and how | CT "with contrast" needs intravascular, intra-articular or intrathecal contrast; oral contrast alone codes as without |
| Contrast sequence | Without, with, or both phases | CT abdomen codes 74150, 74160 and 74170 differ by phase |
| Complete vs limited ultrasound | Every required organ or structure, or why one wasn't seen | A complete abdominal ultrasound (76700) needs the full organ list; otherwise it's limited (76705) |
| Laterality | Right, left or bilateral, consistently | Drives RT, LT or modifier 50 and unilateral vs bilateral codes |
| Guidance | That imaging guided the procedure, and that images were retained | Ultrasound guidance codes require a permanently recorded image |
| 3D rendering | Whether it was done, and on a separate workstation | 76376 vs 76377 depends on that distinction |
Code numbers only. Full descriptors are AMA copyright; check your licensed CPT codebook.
The ultrasound row is where the most revenue is lost. A 2025 coding guide from APS Medical Billing lists the structures a complete abdominal ultrasound must document (liver, gallbladder, common bile duct, pancreas, spleen, kidneys, upper abdominal aorta and inferior vena cava) and notes that a missing organ needs a reason, such as a prior cholecystectomy. A report that says "abdomen unremarkable" without naming the structures supports only the limited code.
Report elements that drive the diagnosis
The diagnosis comes from two parts of the report: the clinical information and the impression.
When the study finds something. The ICD-10-CM Official Guidelines (Section IV.K) say that for outpatient diagnostic tests interpreted by a physician, when the final report is available at the time of coding, you code the confirmed or definitive diagnosis in the interpretation, and don't add the related signs and symptoms.
When the impression is uncertain. Outpatient coding (Section IV.H) doesn't code "probable," "suspected," "likely" or "rule out" diagnoses. You code the sign, symptom or abnormal finding instead. "Findings likely represent pneumonia" codes as the symptom from the indication, not as pneumonia. Radiologists hedge for good clinical reasons, so this is the most common gap between what the radiologist meant and what can be coded. If coders see it often, the answer is template wording, not pressure on radiologists to overcommit.
When the study is normal. The indication carries the claim. A normal chest radiograph with the clinical history "cough and fever for three days" codes the symptoms, which supports medical necessity. A normal study with no history, or "rule out pathology," leaves nothing to code but a routine encounter, which most payers won't pay. That's why the ACR parameter's "relevant clinical information" field matters to revenue.
One related change: the Medicare appropriate use criteria program for advanced imaging no longer applies. CMS paused the program and rescinded its regulations effective January 1, 2024, so AUC consultation details are no longer reported on Medicare claims. The indication still has to support medical necessity under coverage policy.
The signals AI-assisted coding depends on
Radiology is often one of the first settings where AI coding is deployed, because reports are short, templated and closely tied to a small code range. That only holds when the report states the facts above in words the software can find. The direct-to-bill vs coder-in-the-loop decision rests on how often those facts are present.
What well-run systems do when a signal is missing:
| Missing or unclear signal | What a careful system does |
|---|---|
| Contrast stated in the header but not the body | Routes to a coder; header text alone isn't evidence of the service |
| Ultrasound with an incomplete organ list | Codes limited, or routes for review if the template label says "complete" |
| Laterality conflicts between header and body | Routes to a coder; never guesses |
| Uncertain impression with no indication | Routes to a coder or a query; nothing safe to code |
| Addendum after first coding | Re-reviews the claim before release |
| Order and performed study differ | Flags for compliance review |
The same gaps cost a human coder time and cost an automated system its confidence. Improving report templates raises both coder productivity and the share of studies software can code without review. When a vendor quotes a radiology automation rate, ask what share of reports in the sample were template-driven and how missing elements were handled; the methodology for measuring accuracy explains why the denominator matters. MEDIQCODE's approach is AI-assisted coding with human sign-off, and we'd expect any tool to show its reasoning per code so a coder can check it.
A report-quality audit you can run
You don't need software to find out how codable your reports are. Pull 100 finalized reports across your top modalities and score each against ten questions:
- Is the clinical indication stated in signs, symptoms or a diagnosis?
- Does the body name the exam actually performed?
- Is the number of views stated for radiographs?
- Is contrast stated, with route and phase?
- For ultrasound, are all required structures named, or absences explained?
- Is laterality stated and consistent between header and body?
- Is guidance documented with retained images where applicable?
- Does the impression give a definitive diagnosis where the findings support one?
- Are nonroutine communications documented with date, time, method and recipient?
- Do any addenda change a coded element?
Report the failure rate per question and per radiologist template, not per person. Most failures trace to a template default, such as an ultrasound template labeled "complete" that doesn't prompt for each organ. Fix the template, and the next 10,000 reports improve without anyone changing how they dictate. Do this before evaluating any automation tool, because the result becomes the baseline you'll measure it against. For the rest of the evaluation, see how accurate AI medical coding is in 2026.
Frequently asked questions
What are the documentation requirements for a radiology report?
The ACR Practice Parameter for Communication of Diagnostic Imaging Findings recommends demographics, relevant clinical information, the procedures and materials used, findings, limitations, clinical issues, comparison studies, and an impression. For coding, the body must also state views, contrast route and phase, laterality, and the structures examined.
Can a coder use the order or exam header to pick the CPT code?
No. Code from the body of the signed report, which describes what was performed. If the body and the order differ, the report should explain why, and the difference may need compliance review.
How do you code a radiology report that says "likely" or "suspected"?
For outpatient studies, don't code the uncertain diagnosis. Code the signs, symptoms or abnormal findings instead, per Section IV.H of the ICD-10-CM Official Guidelines.
Does Medicare still require appropriate use criteria for advanced imaging?
No. CMS paused the AUC program and rescinded its regulations effective January 1, 2024. Medical necessity still has to be supported by the indication and applicable coverage policy.
This article is general education on documentation and coding practice. It is not a coding determination for any specific study. Code assignment depends on the signed report, current code sets and applicable payer policy.
