On June 25, 2026, Aidoc announced that the FDA granted Breakthrough Device Designation to First Read, AI designed to analyze chest radiographs and generate preliminary radiology report text. For radiologists, the designation signals a faster path toward FDA clearance. For medical coders and CDI specialists, it signals something else: the source document that drives ICD-10-CM diagnosis codes, CPT procedure codes, and HCC capture is about to be drafted by a machine across nearly 2,000 hospitals worldwide.
That shift is worth understanding now, before AI-authored reports become a routine part of your documentation workflow.
What Aidoc’s Breakthrough Device Designation Actually Means
The FDA’s Breakthrough Device Designation (BDD) is granted to technologies that significantly advance the diagnosis of serious conditions and address an unmet clinical need. It accelerates the FDA review process but does not mean the device is cleared or commercially available — First Read remains investigational. Aidoc already deploys clinical AI across nearly 2,000 hospitals and has analyzed more than 120 million patient cases. The company raised a $150 million Series E in April 2026, led by Growth Equity at Goldman Sachs. First Read is Aidoc’s second BDD in under a year, following CARE Triage in September 2025.
First Read is built on Aidoc’s CARE foundation model. It reads a chest radiograph, identifies findings, and generates the preliminary text of a radiology report — the draft the radiologist then reviews and signs. The four indications covered by the BDD are life-threatening findings on chest X-ray. Once the radiologist attests, that signed report enters the medical record and becomes the documentation coders work from.
The Report Is the Record — and the Record Drives Coding
Medical coding is documentation-dependent. A coder cannot assign a more specific diagnosis than the interpreting physician has documented. In radiology, the interpretation report is that documentation.
The difference between J18.9 (pneumonia, unspecified organism) and J13 (pneumonia due to Streptococcus pneumoniae) is entirely a documentation question — not a clinical one. For a Medicare Advantage patient, those codes map to different HCC categories with meaningfully different risk-adjustment weights. For an inpatient case, they can affect DRG assignment. When AI drafts the report, the specificity of that draft determines whether the final signed document contains language precise enough for an accurate code assignment.
If First Read outputs “patchy bilateral opacities consistent with pneumonia” and the radiologist signs without addendum, the coder is left with an unspecified code — not because the clinical picture was unclear, but because the AI’s draft language lacked specificity. The gap is not clinical. It is documentation.
Where AI-Generated Reports Introduce CDI Risk
The Specificity Problem in Foundation Model Output
Foundation models trained on radiology reports tend to reproduce the language patterns of their training data. If the training corpus skews toward generic boilerplate — the kind that has historically populated PACS systems — the AI will reproduce that boilerplate at scale. “Bilateral pleural effusions” is not the same, for coding purposes, as “moderate bilateral pleural effusions with adjacent compressive atelectasis in the setting of known decompensated heart failure.” The second phrase enables a more precise coding chain. The first requires the coder to hunt for supporting context elsewhere in the record, or initiate a CDI query.
CDI specialists reviewing AI-drafted reports will need to apply the same query rigor they use for human-authored notes, probably more. The AI has no clinical relationship with the patient, no memory of prior encounters, and no incentive to surface diagnostic nuances that affect coding accuracy. That burden shifts explicitly to the CDI and coding team.
CPT Billing in the AI-Assisted Reading Room
For now, the radiologist who reviews and signs First Read’s output bills the standard CPT code for the interpretation — for example, CPT 71046 for a two-view chest X-ray. The AI’s contribution is not separately billable under current CMS policy for investigational devices. Once First Read completes FDA clearance and enters commercial deployment, the billing picture could change.
The AMA has been actively expanding CPT codes for AI-augmented imaging. The progression from Category III (temporary) to permanent Category I codes is already documented in other modalities: Cleerly’s coronary plaque AI moved from Category III codes 0623T through 0626T to permanent Category I code CPT 75577, effective January 1, 2026. A similar trajectory for AI-assisted report drafting is plausible as these tools clear FDA review. Coders and billing managers should monitor AMA CPT quarterly updates for any new Category III codes addressing AI-generated radiology report workflows.
What Coding and CDI Teams Should Do Now
Aidoc’s Breakthrough Device Designation is a signal that AI-drafted radiology reports are on a near-term path to clinical deployment. Here is where coding and CDI teams can prepare:
- Map your high-HCC-risk radiology findings. Chest X-ray interpretations touching heart failure, COPD, pleural effusion, and pulmonary embolism map to HCC categories that materially affect Medicare Advantage risk-adjustment revenue. These are the findings where AI specificity gaps will cost the most.
- Update CDI query templates for AI-drafted language patterns. Existing query templates assume human-authored notes. Review early AI-drafted reports at your facility to identify recurring specificity gaps and create targeted query templates before volume scales.
- Clarify physician attestation standards. When a radiologist signs an AI-drafted report, that signature is the physician’s attestation of accuracy. Compliance and coding leadership should confirm with radiology what review depth is required before AI drafts enter the permanent medical record.
- Track CPT Category III updates for AI imaging codes. New codes for AI-assisted interpretation services may appear as First Read and similar tools clear FDA review and enter commercial use.
- Build an audit trail for AI-authored documentation. For OIG audit preparedness, knowing which reports were AI-drafted and which received full radiologist dictation will be material if payers scrutinize ICD-10-CM code specificity on chest X-ray claims.
The Coder’s Expanding Role in AI Oversight
There is a broader implication here. As AI tools draft more of the documentation that coders work from, coders and CDI specialists are no longer just the end of the coding chain. They become a quality-assurance layer on the AI’s output. A coder who identifies a pattern of unspecified respiratory codes tracing back to AI-drafted chest X-ray reports is, in effect, detecting a model performance issue that affects both patient safety documentation and revenue integrity. That is a more strategic role than most coding departments have historically been positioned to play.
Aidoc CEO Elad Walach said in the June 25 announcement: “Radiology is entering a new era.” For medical coding, so is the entire upstream documentation process — and the teams who manage that documentation have more leverage than ever.
Keeping pace with AI-generated documentation is central to accurate, audit-ready coding. Learn more about Medikode’s automated medical coding platform and how it addresses quality in AI-driven workflows.