Healthcare AI Navigator (@HealthAINav) header
Healthcare AI Navigator (@HealthAINav) profile picture

The latest in healthcare AI. Insights, breaking news, journal findings & real-world applications for clinicians and healthcare leaders. #HealthcareAI

Joined October 2026
Photo by Healthcare AI Navigator (@HealthAINav)Photo by Healthcare AI Navigator (@HealthAINav)Photo by Healthcare AI Navigator (@HealthAINav)Photo by Healthcare AI Navigator (@HealthAINav)Photo by Healthcare AI Navigator (@HealthAINav)Photo by Healthcare AI Navigator (@HealthAINav)Photo by Healthcare AI Navigator (@HealthAINav)Photo by Healthcare AI Navigator (@HealthAINav)Photo by Healthcare AI Navigator (@HealthAINav)Photo by Healthcare AI Navigator (@HealthAINav)
A big week in healthcare AI. Here are 5 developments worth knowing about: 1. CMS: Released a roadmap exploring how AI could be regulated and paid for in clinical care. It's a plan, not a new rule. 2. Epic: Authorized external AI agents will be able to retrieve EHR data via MCP, with connections planned for Q1 2027. 3. OpenAI: Announced new computer-use capabilities for healthcare workflows and 20+ healthcare plugins. 4. Apple + OpenEvidence: Launched a US study inviting up to 200,000 participants to explore how people use AI for health questions. 5. Isomorphic Labs: Raised $5 billion to advance AI-designed drugs toward clinical trials. Huge investment, but clinical success is still to be demonstrated. The theme this week? AI is moving beyond answering questions and toward actually doing things across healthcare. That brings some pretty big opportunities, and some equally big questions about safety and oversight. #healthcareAI #digitalhealth
4
1
1
25
1,693
Could a smartphone photo help surgeons plan facial reconstruction without another CT scan? Researchers have developed Tele-Cranio, an AI system that uses 2D facial photos to create 3D models of the face, estimate underlying bone structure and preview possible postoperative appearance based on a surgeon's plan. It was developed using 539 paired samples, including data from 334 patients with craniofacial deformities, spanning ages 3 to 60. The potential is in remote assessments, early consultations and follow-up, particularly where repeated imaging or travel is difficult. But this is research-stage technology. The authors explicitly position it as a complement to CT, not a replacement for definitive diagnosis or surgical planning. Source: npj Digital Medicine, Nov 20, 2026. #HealthcareAI #HealthcareInnovation
53
More than 5 million patients' records could underpin a new AI research platform at Cedars-Sinai. The hospital and nference have announced a long-term collaboration to analyze de-identified clinical data spanning more than a decade. The planned platform will bring together notes, lab results, imaging, genomic data and other sources so researchers can study diseases across multiple organ systems and stages of care. The goal is to support research, clinical trial design and development of new AI models. It isn't a newly deployed diagnostic tool or evidence of better patient outcomes. There's considerable promise in longitudinal data, but the value will depend on the quality of the data, safeguards and what the research actually delivers. Source: nference, Nov 20, 2026.
1
2
218
A microscope that helps guide its own experiment? Danaher and Microsoft want to make that part of laboratory work. Their newly announced partnership is developing AI agents that can translate scientists' instructions into experiment plans, assist with instrument setup, interpret results and suggest next steps, with human oversight. In preliminary internal testing, novice researchers using an AI assistant on a Leica microscope completed one workflow in up to 83% less time and up to 68% fewer steps than novices without it. That's a company-reported result from a specific task, not proof that whole laboratories will become 83% faster. A first customer-facing assistant is expected in 2026, with additional agent-enabled instruments planned for 2027. Source: Danaher, Nov 20, 2026.
1
148
A Stanford study suggests AI could take a small but persistent documentation task off operating room nurses' plates. Using ambient video and computer vision, an AI system automatically recorded when patients entered and left the OR. Researchers evaluated it across 14 operating rooms and one procedure room and found its timestamp accuracy was not inferior to manual documentation. These timestamps feed into scheduling, staffing and room-turnover decisions. But this was a single-center validation study, not proof that hospitals save staff hours or improve patient outcomes. It's an interesting reminder that useful healthcare AI doesn't always need to diagnose disease. Sometimes it just needs to capture routine information reliably, without asking a busy clinician to stop and type. Source: JAMA Surgery, November 18, 2026.
85
AI reduced the number of patient records needing manual review for a simulated cancer trial by approximately 89%. Fujitsu Japan, National Cancer Center Hospital East and Paradigm tested an AI-assisted screening workflow using data from 22,334 respiratory patients. For an EGFR-mutated lung cancer trial scenario, the system prioritized 57 potential participants. After reviewing the supporting records, physicians judged 51 to be promising candidates. The 89% figure measures reduced chart-review workload, not faster enrollment or improved patient outcomes. This was a verification study, and the service is not yet commercially launched in Japan. Trial recruitment often depends on staff finding the right details buried in records. This is a promising use of AI, provided clinicians still make the final eligibility decisions. Source: Fujitsu Japan, November 19, 2026.
78
Epic says authorized AI agents built outside its EHR will be able to retrieve patient-record data using a new connection designed for AI. At HLTH this week, Epic outlined plans to support the Model Context Protocol (MCP) by the first quarter of 2027. It also says its own agents will be able to connect with external applications. Epic reports more than 3,700 apps already connected to its platform. But the announcement arrives amid scrutiny of how much access competing developers actually get. The new capability is planned, not broadly available today, and details of which records outside agents can retrieve are still limited. For hospitals choosing AI tools, access to EHR data may matter as much as the AI itself. Who gets to connect, and on whose terms? Source: Epic and STAT, November 18, 2026.
6
13
3
101
10,197
An AI answer can be clinically accurate and still cite research that doesn't exist. In a new study, two urologists assessed ChatGPT 5.2 Pro and Gemini 3 Pro on 65 questions about kidney stones, using European clinical guidelines. Gemini's answers were rated clinically appropriate in 98.5% of cases, compared with 81.5% for ChatGPT. But 45.8% of ChatGPT's individual citations and 24.8% of Gemini's were entirely fabricated. Another 30.3% and 31.4%, respectively, pointed to real papers with incorrect or mismatched DOIs. There's an important caveat: researchers specifically required 3–5 detailed references per answer, and tested each question just once. These figures aren't general error rates for all healthcare AI use. For clinicians and researchers, the takeaway is simple: a medically convincing answer doesn't guarantee the supporting papers are real. Source: Scientific Reports, Nov. 18, 2026.
91
💰 A massive week for healthcare AI investment. $5 BILLION. That's how much Alphabet-backed Isomorphic Labs just raised to develop AI-designed medicines. The funding round, led by Greenoaks and Sequoia Capital, will help advance its drug discovery technology and move potential treatments toward clinical trials. Meanwhile, investment is also flowing into clinical decision support, medical imaging AI validation, and tools that make patient data more useful. The contrast is striking: billions going toward discovering tomorrow's medicines, while smaller investments focus on improving how healthcare works today. The big test? Whether those investments translate into better outcomes for patients.
1
1
118
AI is moving directly onto bedside X-ray machines, without needing a cloud connection. GE HealthCare has announced Critical Care Suite 3.0, FDA-cleared software that detects feeding tubes on portable X-rays and marks their path, tip and side port. The feeding tube assessment now covers both adults and children. In GE's testing, the AI achieved 99.2% sensitivity for tube detection and 91.6% mean DICE localization, measuring how closely its outlines matched expert annotations. For ICU and emergency teams, having this information available at the bedside could be useful. But these are technical performance results, not proof that the software prevents complications or replaces clinical review. Source: GE HealthCare, November 17, 2026.
1
65
Oracle has launched AI agents designed to help manage clinical trials, from planning studies to identifying operational problems. The new Life Sciences AI Agents bring together information from clinical trial systems to monitor study performance, flag emerging risks and recommend next steps. For example, if patient recruitment is falling behind schedule, an agent could identify the issue, investigate possible causes and suggest actions for the research team. Importantly, these agents aren't intended to make decisions independently. Humans must review and approve proposed actions before they're carried out. The technology is now available to early adopters, initially focusing on trial planning, monitoring and operations. Oracle hasn't yet reported independently validated improvements in trial costs, timelines or outcomes. For research teams juggling multiple sites, protocols and deadlines, this could be a useful application of agentic AI. But its real value will depend on how reliably it identifies problems and whether its recommendations actually improve trial performance. Source: Oracle, November 18, 2026.
1
34
An AI-powered Fitbit coach got people walking more. But was it much better than ordinary reminders? In a randomized trial, 13,463 Fitbit users received either AI-personalized activity prompts, fixed prompts, random prompts or no prompts. The main analysis included 7,711 participants. After two months, the AI group gained about 210 more daily steps than the no-reminder group. That's statistically significant, but modest. By then, the AI approach didn't significantly outperform either of the simpler reminder strategies. The trial only followed people for two months, and most of those analyzed were women. For digital health tools, maybe the most useful comparison isn't AI versus doing nothing. It's whether AI delivers enough extra benefit to justify the added complexity. Source: Journal of Medical Internet Research, Nov 17, 2026.
1
1
51
Medicare's national fee-schedule amount for an AI diabetic eye exam is $46.76. But who makes sure patients with a positive result get the care they need? A new MedCity News analysis argues that reimbursement has focused on screening while follow-up remains a weak link. In a Houston tele-retinal screening study, just 54.9% of 1,695 patients referred after abnormal results attended an eye specialist appointment. That research did not evaluate autonomous AI. The challenge isn't just making an accurate detection. It's booking the referral, reaching patients and having specialist capacity. If we count a completed scan as success, we risk missing the outcome that really matters: whether someone with potential sight-threatening disease actually receives care. Source: MedCity News, Nov 17, 2026; International Ophthalmology Clinics, 2025.
1
1
1
42