Last week, I was struck by FDA approval of software that generates a ready-to-edit report for the radiologist from ultrasounds. (Here).
I asked Chat GPT to do a recurrent comples search every Monday.
1) Track FDA K260303 webpage for any updates like release of safety & effectiveness review.
2) Track trade journals for articles on auto-generation of reports, in radiology & pathology
3) Track PUBMED for peer review articles on auto-generation of reports, in radiology & pathology
Weekly AI-generated-report watch
FDA K260303 — important baseline update. The FDA record for See-Mode Augmented Reporting Tool, Breast (SMART-B) now contains a live 510(k) Summary and says the database page was last updated August 10, 2026. However, I do not find a separately posted FDA Decision Summary or Safety and Effectiveness review/SSED. The available document is the sponsor’s 24-page 510(k) Summary, not an FDA-authored review memorandum.
The document is nevertheless unusually relevant to our question. It states explicitly that SMART-B analyzes breast-ultrasound images, generates BI-RADS categories and descriptors, and then “generates a report based on the image analysis results” for physician review, modification, and approval. So the regulatory record now gives us quite clear documentary evidence of FDA clearance of an image-analysis → machine-generated draft report → physician QC/finalization workflow.
Trade literature, August 3–10. One item is worth noting, although it concerns AI-generated summaries rather than autonomous diagnostic reports: Radiology Business/Health Imaging, August 5, 2026, reported on a new ACR resource for patients using AI-generated summaries of radiology reports. It is evidence that generated radiology text is moving into practical workflow and patient communication, but I would classify it as adjacent rather than another SMART-B-type image-to-report system.
A second useful signal appeared in The Pathologist on August 6, 2026. In an interview on AI in biomarker assessment, Roche’s Audrey Bennett said the near-term computational-pathology trajectory remains collaborative rather than autonomous, with pathologist judgment retained and regulatory attention focused on the end-to-end diagnostic system. That is pertinent to our pathology-report thesis, but it does not describe a newly deployed machine-generated pathology-report product.
PubMed. I found no PubMed-indexed paper newly published during August 3–10 that meets the strict criterion of substantive progress toward machine generation of diagnostic radiology or pathology reports. Searches did identify recent 2026 work on radiology-report generation and pathology vision-language models, but the directly relevant papers were published or placed online earlier than this seven-day window. For example, the recent American Journal of Pathology review explicitly identifies report generation as an emerging pathology VLM capability, but it was accepted July 1 and published online in July, so I am not counting it as this week’s development. PMID 42508637 Xia et al.
Bottom line: no new FDA Decision Summary/Safety-and-Effectiveness document yet. The most important development is that the current K260303 FDA record now exposes the detailed 510(k) Summary, which very plainly documents the machine-generated-draft-report architecture we have been tracking. I’ll continue watching specifically for the separate FDA review documents and for genuinely new image-to-report systems rather than merely generic generative-AI or report-simplification articles.