AI in Radiology Practices

A radiology AI startup founder just walked us through how they’re preparing their FDA submission. 47 pages on model architecture, ROC curves, sensitivity/specificity tables, and test datasets. Zero pages on how a radiologist actually uses it between reads. This disconnect is why so many AI/ML medical devices face friction and delays in FDA review, even when the underlying model is strong. We saw the opposite pattern with BodyCheck, a cardiovascular screening tool built on chest X-rays. When we started working with their team, the critical risk wasn't whether the AI could measure cardiothoracic ratio (CTR). It was whether radiologists would actually use it at 3 p.m. on a Tuesday with 40 studies still in queue. Two themes kept coming up in conversations and feedback: • Radiologists didn’t want to click into a separate app just to see AI results. • They wanted the system to surface when it wasn’t confident, so they’d know when to ignore it. So instead of building for a conference demo, the product was built for the reading room: • Results delivered directly into existing PACS workflow, not a standalone dashboard • DICOM viewer with overlays right on the chest X-ray • Clear CTR value + normal/abnormal flag, not a heatmap buried in another • Confidence scores surfaced explicitly, so radiologists know when the algorithm encounters unusual anatomy or low-quality images The "feature list" looked dull compared to flashy AI platforms: • No 23-feature analytics suite. • No admin-facing "command center." But it did the only thing that matters: it fit seamlessly into a radiologist's real day. While many AI imaging tools remain stuck in pilot purgatory, BodyCheck went from research project to FDA-cleared SaMD with real clinical adoption: • Algorithm developed and productionized (MONAI → TensorFlow) • Integrated directly with PACS for automatic image retrieval and result distribution • 510(k) submitted in 3 months, cleared in ~15 weeks Why the fast clearance? The design and documentation told a coherent story about real clinical use, not just isolated technical capability. When we debriefed with users, they said: • "I don't have to change how I read." • "It tells me when not to trust it." • "It just shows up where I already work." Shadow a radiologist to watch how many tools get "piloted" then quietly ignored. The pattern we keep seeing in health tech: Products that actually make it into daily workflow obsess over one painful moment in a clinician's day and quietly make it disappear. The solution that wins isn't the one with the strongest metrics on paper, it's the one that shows up in the radiologist's existing viewer and just works. Learn how we help AI medical device teams navigate FDA clearance: https://hubs.li/Q03VL-PF0 #FDA #510k #AIinHealthcare #Radiology #DigitalHealth #SaMD #HealthTech #MedTech #RegulatoryStrategy #RSNA #RSNA2025

Workflow Wednesday: Why Integration Matters More Than Innovation In healthcare, we often get dazzled by shiny new technologies. A new algorithm that detects disease faster. A new model that beats human performance in a study. But here’s the truth I’ve learned as both a radiologist and as Chief Medical Officer at GE HealthCare: the real bottleneck is rarely the algorithm. It’s the workflow. A brilliant #AI tool that sits outside the daily routine of a radiologist won’t move the needle. If it adds clicks, slows reporting, or disrupts communication with clinicians, it risks becoming just another “nice idea” that never scales. This is why integration is everything: AI must live inside the worklist, not in a separate portal. It must prioritize urgent cases automatically, not wait for manual triage. It must speak the language of radiology reports, not produce outputs no one knows how to use. The most successful innovations I’ve seen are not the flashiest. They are the ones that disappear into the workflow so seamlessly that clinicians almost forget they’re using AI at all. As leaders, we need to remember: technology succeeds not when it looks impressive in a demo, but when it quietly improves care at scale. Question for you: In your own practice, what’s the biggest barrier to integrating new tools: technology itself, hospital IT, or cultural resistance? #Radiology #AIinHealthcare #Leadership #gehealthcare #WorkflowWednesday

🌟 This editorial from Radiology by Merel Huisman, MD, PhD, Felipe Kitamura, MD, PhD, Tessa Cook, Keith Hentel, Jonathan Elias, George Shih, and Linda Moy discusses the benefits and challenges of using large language models (LLMs) in clinical radiology, specifically focusing on clinical decision support, society guidelines and best practices, accuracy monitoring, academic administrative support, open-source and commercial LLMs, and agentic workflows. 🔍 It explores the potential of LLMs to enhance radiologists' work, highlighting their capabilities in generating reports, improving diagnostic accuracy, and providing patient-centered information. 🚨 The authors warn against overreliance on LLMs and the need for continuous monitoring, emphasizing the importance of maintaining accuracy and addressing biases. They advocate for combining quantitative metrics with qualitative user feedback. 💯 🤖 The article also explores the development of open-source LLMs, a potential solution to avoid overdependence on commercial LLMs, and the emerging field of agentic workflows, where LLMs can perform tasks and make decisions autonomously. 👍🏼 It's refreshing to see radiology finally discuss agentic AI. Overall, this editorial provides excellent insight into LLMs in radiology, highlighting not only their potential but also their pitfalls and limitations. It's a recommended read for anyone interested in LLMs in radiology. 🔗 to the editorial is in the first comment. 👇🏼 #Radiology #ArtificialIntelligence #LLMs #GenAI #AgenticAI Radiological Society of North America (RSNA) #RadiologyAI

One AI just replaced 14 radiology tools. Game changer or hype? Yesterday changed radiology forever. The FDA cleared Aidoc's CARE foundation model, the first AI to detect 14 acute conditions from a single abdominal CT scan. Think about what this means: Instead of running separate AI tools for appendicitis, bowel obstruction, liver injury, spleen trauma, and 10 other conditions, radiologists now have ONE system that catches everything. 📊 The numbers are staggering: • 97% sensitivity (up to 98.5%) • 98% specificity (up to 99.7%) • 10x fewer false alerts than single-condition tools • 100 million patient cases already analyzed But here's what really matters: Emergency departments are drowning. Imaging backlogs are killing patient flow. Radiologists are reading scans on a first-come, first-served basis while critical cases sit in queue. This AI changes that equation. It pulls urgent findings to the front of the line. Automatically. A patient with internal bleeding doesn't wait behind 20 stable cases anymore. Yet the real breakthrough isn't the technology, it's the approach. For years, we've been bolting on single-purpose AI tools like adding apps to a phone. Each one solved one problem. Integration was a nightmare. Alert fatigue was real. Aidoc just proved foundation models work in clinical practice. One model. Multiple conditions. Better accuracy. This is how AI actually scales in healthcare: not through hundreds of narrow tools, but through comprehensive systems that match how clinicians think. The question isn't whether this transforms emergency radiology. It's whether other specialties are paying attention. Because if a single AI can master 14 acute abdominal conditions, what's stopping it from learning chest, brain, or musculoskeletal imaging next? We just witnessed the iPhone moment for medical AI. The consolidation has begun. ♻️ Repost if emergency departments need AI triage yesterday 👉 Follow me, Jonathan Govette, for daily, real-time updates on healthcare technology and business news. LinkedIn Profile: https://lnkd.in/gWyNQkDn

Could AI drafts—even imperfect ones—be a time-saver for radiologists when interpreting CT scans? Our pilot study using simulated AI reports found a 24% faster workflow, with accuracy intact. Q: What makes this study's approach unique? A: Instead of building an AI system, we used GPT-4 to simulate what AI-generated draft reports might look like. We deliberately introduced 1-3 errors in half the drafts to study how radiologists would handle imperfect AI assistance - a "Wizard of Oz" approach to prototype the future workflow. Q: How was the simulation study structured? A: We conducted a 3-reader crossover study with 20 chest CT cases. Each case was read twice: once with standard templates, and once with our simulated AI drafts. This controlled design let us directly compare the workflows. Q: What efficiency gains did you see with the simulated drafts? A: Median reporting time dropped from 573 to 435 seconds (p=0.003) - a 24% reduction. Two readers showed major improvements (717→398s and 361→322s), while one showed an increase (947→1015s). Q: Did the intentionally flawed drafts impact accuracy? A: Surprisingly, even with deliberately introduced errors in half the simulated drafts, the AI-assisted workflow showed slightly fewer clinically significant errors (0.27±0.52) compared to standard workflow (0.38±0.78). While not statistically significant, this suggests radiologists maintained their vigilance even with imperfect drafts. Q: How did radiologists respond to working with these simulated drafts? A: All 3 readers found the prototype system easy to use and well-integrated into their workflow. Two reported somewhat less mental effort, while one reported significantly reduced effort. Their likelihood to recommend it varied (scores of 5, 9, and 10 out of 10). Q: What's next? A: While these simulation results are encouraging, these are small scale pilot studies setting the stage for deeper validation. Link to short paper: https://lnkd.in/d-4aTJ69 Congratulations to stellar team of Julián Nicolás Acosta, Siddhant Dogra, Subathra Adithan, Kay Wu, MD 💫, Michael Moritz, Stephen Kwak

🔥 𝗥𝗲𝗮𝗱 𝗮𝗻𝗱 𝘀𝗵𝗮𝗿𝗲 𝘁𝗵𝗶𝘀 𝗽𝗮𝗽𝗲𝗿! The results are surprising: "Performance gains from 'AI + clinician' are not automatic; rather, they depend on multiple factors, including task attributes, interface and prompt design, human-factors training, and workflow integration." 𝗥𝗲𝘀𝘂𝗹𝘁𝘀 "Four themes—the collaboration paradox, task specificity, quality–accuracy trade-offs, and the primacy of human factors–collectively indicate that merely placing an AI tool in clinicians’ hands does not guarantee consistent net benefit." "A critical finding...is the 'collaboration paradox': H+AI collaboration does not demonstrate universal superiority over a strong AI-only agent." "...H+AI accuracy (58%) did not significantly exceed, and was directionally similar to, standalone AI accuracy (~60%)." "As noted in our sensitivity analysis, the estimated synergy ratio for this task was approximately 1, providing no statistical evidence of true collaborative gain over the AI only agent." "Plausible mechanisms include cognitive dissonance when AI advice conflicts with clinicians’ initial judgments and miscalibration driven by automation and confirmation biases." 𝗣𝗿𝗮𝗰𝘁𝗶𝗰𝗮𝗹 𝗜𝗺𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻𝘀 "This indicates that 'human-in-the-loop' is not a universal safeguard, but a complex interaction that can introduce new failure modes—such as anchoring on incorrect AI suggestions or diluting high-performing AI outputs through inconsistent human overrides." "Consequently, workflow implementation should move beyond a one-size-fits-all collaboration model toward a task-differentiated strategy." "For highly structured, low-variance tasks where AI performance is robustly superior, strategic delegation to AI-only pathways with targeted 'human-on-exception' oversight may be optimal." "In contrast, for high-ambiguity, high-stakes scenarios (e.g., complex differential diagnosis), workflows must be deliberately engineered to support effective human–AI teaming, with interfaces that surface model uncertainty, provide contrastive and traceable evidence chains, and embed mandatory verification protocols or secondary sign-offs." 𝗔𝗽𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻𝘀 "To guide deployment, we propose a 2 × 2 classification framework defined by task complexity (high/low) and task structure (high/low): collaborative value is greatest in the high-complexity, low-structure quadrant (e.g., ambiguous presentations with broad differentials) and minimal in the low-complexity, high-structure quadrant (e.g., templated documentation)." "...clinicians should shift from 'information generators' to expert verifiers, with core competencies in critically appraising and recalibrating AI outputs." 𝗧𝗵𝗲 𝗕𝗼𝘁𝘁𝗼𝗺 𝗟𝗶𝗻𝗲 "Ethically and legally, ultimate 𝘳𝘦𝘴𝘱𝘰𝘯𝘴𝘪𝘣𝘪𝘭𝘪𝘵𝘺 𝘳𝘦𝘮𝘢𝘪𝘯𝘴 𝘸𝘪𝘵𝘩 𝘤𝘭𝘪𝘯𝘪𝘤𝘪𝘢𝘯𝘴 despite AI assistance; this supervisory and accountability burden (“vigilance tax”) must be supported by governance, audit, and traceability mechanisms."

Generative AI isn’t replacing radiologists but may soon assist like a well-trained resident. 🩻 👇 A new Nature Perspective presents the frontier of Multimodal Generative AI (GenMI) in healthcare. This new class of AI models is not just interpreting medical images, it’s generating narrative reports, integrating clinical history, and even offering real-time interaction with clinicians and patients. The paper calls for a shift from single-task automation to holistic, collaborative AI assistants, or what the authors term the “AI Resident.” 👉 Key Takeaways 1. Beyond Detection: Toward Narrative Intelligence GenMI models go beyond triaging or highlighting findings,they synthesize multimodal data (e.g., imaging + clinical history) into coherent, structured reports that can rival expert drafts. 2. The “AI Resident” Paradigm Envisioned as a collaborative tool, the AI resident supports clinicians in drafting reports, enables interactive querying of findings, and can even assist in patient education and trainee feedback loops. 3. Multimodal & Multispecialty Applications While radiology is the focal domain, GenMI is expanding into pathology, dermatology, ophthalmology, and endoscopy, powered by vision-language models like GPT-4V and Google’s Gemini. 4. Challenges: Bias, Hallucination & Evaluation Gaps GenMI systems are prone to hallucinations and performance drops across underrepresented populations. Traditional NLP metrics are inadequate; new benchmarks like RadBench and RadGraph F1 are being proposed. 5. A Call for Responsible Deployment Authors advocate for gradual clinical integration, open benchmarks, diverse datasets, and human-in-the-loop calibration to ensure GenMI complements not replaces expert judgment. 🎯 GenMI represents a pivotal evolution in clinical AI from task-specific tools to interactive, multimodal assistants. If deployed with care, the AI resident could reduce burnout, democratize expertise, and reshape how medical knowledge is generated, shared, and acted upon. _________________________________________________________ #radiology #machinelearning #ai #medicine #health

The Generative AI Healthcare Playbook: Where to Invest Now vs. Later Generative AI is revolutionizing healthcare across multiple fronts, with applications maturing at different rates and offering varying levels of immediate value versus long-term potential. In high-maturity categories, AI-powered medical imaging solutions like Aidoc and Zebra Medical Vision are already FDA-cleared, demonstrating 20-30% improvements in radiology workflow efficiency while reducing diagnostic errors by up to 15%. Documentation automation tools such as Abridge and Nuance DAX are showing even faster adoption, with health systems reporting 50% reductions in time spent on clinical notes and $15,000 annual productivity gains per physician. Virtual health assistants have reached widespread implementation, with leading providers like Mayo Clinic deploying AI chatbots that handle 40% of routine patient inquiries, though these systems still require physician oversight for complex cases. The mid-term horizon (2-5 years) presents even more transformative opportunities in drug development and personalized care. In pharmaceutical R&D, generative AI is shortening drug discovery timelines from 5 years to as little as 18 months, with Insilico Medicine's AI-designed fibrosis drug entering Phase II trials in record time. Precision medicine applications are achieving 25-35% better treatment response rates in oncology pilots at institutions like MD Anderson, though broader implementation requires solving data integration challenges across EHR systems. Synthetic data generation is accelerating research while addressing privacy concerns, with companies like Syntegra creating datasets that maintain 98% statistical accuracy without using real patient records. Longer-term innovations show promise but face significant development hurdles. AI-optimized prosthetics are achieving 15-20% better mobility outcomes in trials but require FDA approval and manufacturing scale-up. Drug repurposing algorithms identified baricitinib as a COVID-19 treatment 6 months faster than conventional methods, yet most findings still need clinical validation. For healthcare executives, this landscape demands a balanced investment strategy: implementing proven imaging and documentation AI for immediate 12-18 month payback periods, while allocating 15-20% of innovation budgets to build capabilities in drug discovery and personalized medicine. Critical to success will be establishing AI governance frameworks that address regulatory compliance (particularly for FDA Class II/III devices), mitigate algorithmic bias through diverse training datasets, and ensure seamless integration with existing clinical workflows through API-enabled EHR connections. Organizations that adopt this dual-track approach—combining quick wins with strategic bets—will be best positioned to capitalize on generative AI's potential in healthcare while managing its implementation risks.

The largest randomized #clinicaltrial on #Healthcare #AI to date (100,000 women) compared AI-supported #mammography screening to standard double reading. Key findings: * AI-supported workflow increased cancer detection by 29%. * The use of AI led to a 44% reduction in the screen-reading workload for radiologists. * AI-supported screening reduced "interval cancers" (cancers diagnosed in the two years between screenings) by 12%. * When cancers were found, they were less aggressive, with 27% fewer aggressive subtypes and 16% fewer invasive cancers compared to the control group. Jan 2026 publication: https://lnkd.in/gnPmdeeV March 2025 publication: https://lnkd.in/g7CypJFH #ai #healthcareai #healthai #research #cancer

Key Factors for Implementing AI-Enhanced Clinical Information Systems Successfully Based on industry reports and first-hand observations, it's evident that the integration of Artificial Intelligence (AI) into healthcare is well underway. However, the gap between a successful implementation and an unsuccessful one can be wide. Here are some key factors that maximize the probability of a successful implementation of an AI-enhanced clinical information system. Key Factors for Successful Implementation: 1. Organizational Leadership, Commitment, and Vision 👓 : Leadership buy-in is crucial. A clear organizational strategy for AI needs to be in place to guide the implementation process. 2. Improving Clinical Processes and Patient Care 👩⚕️ : The end goal should be better patient outcomes. Make sure the AI system aligns with this objective. 3. Involving Clinicians in Design and Modification 💻 : Those who will use the system should have input into its design. This ensures relevance and encourages adoption. 4. Maintaining or Improving Clinical Productivity 📈 : The new system should not disrupt workflow. Ideally, it should increase efficiency, perhaps by automating routine tasks. 5. Building Momentum and Support Among Clinicians 🌟 : Early wins can build momentum. Open communication and training are key for securing clinician support. A 🏥 Vignette: Radiology at Hospital X vs Hospital Y ✅ Hospital X: Dr. Smith, head of radiology, involved her team from the start. They pinpointed specific challenges that AI could address. The result: diagnostic accuracy improved, and image reading time dropped by 25%. The department's capacity increased, patient wait times fell, and the team's initial skepticism turned into strong support for the AI system. ⛔ Hospital Y: In contrast, Hospital Y’s administration relied on an external committee with no clinical experience. Dr. Johnson, a senior radiologist, felt sidelined. The system generated multiple false positives, creating bottlenecks and reducing efficiency. The morale dropped, and the project was ultimately abandoned. These contrasting stories underline the importance of each key factor in implementing AI-enhanced clinical systems. Hospital X succeeded due to its thoughtful approach, while Hospital Y serves as a cautionary tale of what can go wrong when these factors are ignored. #HealthcareAI #ClinicalInformatics #Leadership #PatientCare #ImplementationSuccess