The Draft Comes First: Radiology Partners, Vision-Language Models and the New Economics of the Read
Dr. Dereck Mush, MD, MBA
In partnership with
You know the problem. Now build the fix.
You've spent years close to a broken industry or process. You know exactly what would fix it. But you’ve never had the time or resources to make it happen.
With Lovable, your expertise is all you need. Describe the solution to your problem, a website, a dashboard, an internal tool, and get professional-grade software back. You make every call on how it works, because you're the one who knows what it needs to do.
Start today and take $5 off your first month.
Discount of $5 applies to first month of any paid plan. You need to sign up via this webpage in order to claim the discount. The discount applies to new and returning users without an active paid plan. Terms apply.
CLINICAL AI
1. Radiology Partners' Report-Drafting AI Lifted Detection About 40% Across 237,905 Exams
Radiology Partners' Mosaic reports roughly 40% more expert-confirmed findings in an internal study of 237,905 imaging exams using investigational AI report-drafting models. Image: Mosaic Clinical Technologies.
On November 11, Mosaic Clinical Technologies, a subsidiary of Radiology Partners, reported results from an ongoing internal study of investigational vision-language models developed by its Cognita Imaging AI lab. Across 50 conditions on CT and X-ray, AI-generated draft reports surfaced additional findings confirmed by blinded expert radiologists, improving detection by roughly 40% on average and up to 110% for certain findings.
The details:
Detection: The retrospective study covered 237,905 exams. Enhanced detection averaged 40.9% on CT, 43.4% on chest X-ray and 23.9% on other X-rays. Gains included 73.4% for fractures on head CT and 57.5% for lung nodules or masses.
Accuracy and speed: Some AI drafts had fewer errors than original reports. On chest X-rays, 91.2% of AI drafts were error-free versus 83.0% of radiologist-only reports. Experienced radiologists read CTs more than 27% faster and X-rays 31% faster.
Scale: By October 22, the models had drafted nearly 8 million exams, exceeding 37,000 daily. They remain investigational under a research protocol supporting FDA authorization. The chest X-ray model received Breakthrough Device designation in March.
The Signal: Most radiology AI flags individual findings. Mosaic's models draft entire reports, shifting radiologists from authors to editors. Radiology Partners controls the technology, clinical workflow and imaging volume needed to test that approach at scale. But the results are company-run and not yet peer reviewed. The bigger test is whether detecting more findings improves patient outcomes without triggering unnecessary follow-up imaging.
NEUROTECH
2. Neuralink's New PRIME2 Study Brings Its Brain Implant to Chronic Ischemic Stroke
Neuralink's N1 brain-computer interface will be evaluated in the newly registered PRIME2 study, expanding eligibility to adults with chronic ischemic stroke. Image: Neuralink.
On November 10, Neuralink registered PRIME2 (NCT07929792), an early feasibility study of its N1 brain-computer interface implanted by the R1 surgical robot. For the first time in a U.S. Neuralink registry entry, adults with chronic ischemic stroke and one-sided arm weakness are eligible, alongside people with spinal cord injury or ALS.
The details:
Study design: Up to 10 adults across five U.S. sites, with the University of Miami listed first. The implant translates motor-cortex signals into computer cursor and virtual keyboard control.
Stroke eligibility: Participants must be at least 12 months post-stroke, have weakness in their previously dominant arm and show substantially impaired hand dexterity on the Nine-Hole Peg Test.
Endpoints: The primary measure is serious device-related adverse effects through 90 days. Secondary outcomes include cursor control, independent digital activities and patient-reported experience. Primary completion is expected in November 2027.
The Signal: Chronic stroke could open a much larger market for brain-computer interfaces, but the commercial challenge is tougher. Unlike many patients with severe paralysis, stroke survivors may retain alternative ways to interact with technology. Neuralink must demonstrate meaningful improvements in daily independence, not just cursor accuracy. That distinction will shape clinical adoption and reimbursement.
PRECISION ONCOLOGY
3. Guardant Adds Methylation Predictors to Its Blood Test for Breast and Lung Cancer Therapy Choices
Guardant Health introduces five epigenomic applications for Guardant360 Liquid, expanding blood-based insights for breast and lung cancer treatment decisions. Image: Guardant Health.
On November 11, Guardant Health introduced five epigenomic applications for Guardant360 Liquid, powered by its InfinityAI platform. The technology analyzes methylation patterns in circulating tumor DNA to predict treatment response, extending liquid biopsy beyond identifying actionable mutations.
The details:
Five applications: Predicting SERD benefit in breast cancer, immunotherapy response in lung cancer, blood-based PD-L1 status, additional ALK fusions and improved MTAP deletion detection for clinical trial matching.
Evidence: Retrospective analyses found associations between high methylation scores and longer treatment duration. The breast cancer cohort included 765 patients; the lung cancer cohort included 695, with survival and treatment-duration signals favoring higher scores.
Commercial momentum: Guardant reported Q3 revenue of $361 million (+36%) and 117,000 oncology tests (+58%). It raised 2026 revenue guidance to $1.36-$1.37 billion.
The Signal: Liquid biopsy has largely focused on identifying mutations. Methylation adds information about tumor biology and potential treatment response, increasing the value of an existing blood draw. For Guardant, that creates new clinical and commercial opportunities without requiring a separate test. But prospective validation and payer acceptance remain essential before these scores routinely influence treatment selection.
Related reading
Dr. Dereck Mush, MD, MBA
The Draft Is the Product: What AI ECHO INSIGHT Changes for Echo Labs
Dr. Dereck Mush, MD, MBA
AI Colonoscopy Tools Now Have a 334,000 Procedure Study Behind Them. What It Actually Shows
Dr. Dereck Mush, MD, MBA
The Great Unbundling: What the CVS Split Says About Vertical Integration in Healthcare
Dr. Dereck Mush, MD, MBAKeep exploring