SIIM 2026 AI Landscape: AI Moves Into Imaging Workflow

Good morning, there. At SIIM 2026, 49 of 72 radiology-related sessions focused on AI, representing 68.1% of the analyzed programme.

I analyzed the SIIM 2026 session programme to understand how AI is being positioned across imaging informatics. The strongest signal is that AI is no longer just a standalone model topic. It is becoming part of workflow, enterprise imaging, standards, governance, education, and deployment infrastructure. SIIM shows where radiology AI is heading after the model performance phase, into the operational layer of imaging.

Here's what you need to know about Radiology AI last week:

• SIIM 2026 AI Landscape Shows AI Moving Into Workflow and Infrastructure

• Prostate MRI AI more than doubled PSA specificity

• ICH AI shows the cost of false positives

• Breast ultrasound AI boosted reader accuracy

• Plus: 1 newly released dataset, 6 FDA approved devices & 4 new papers.

📊 SIIM 2026 AI Landscape Shows AI Moving Into Workflow and Infrastructure

RadAISlice: Following our previous ECR analyses, we examined the SIIM 2026 programme to understand how AI is appearing across sessions, formats, learning topics, themes, and speaker roles.

The details:

• 72 radiology-related sessions were included in the analysis.

• 49 sessions focused on AI, representing 68.1% of the analyzed programme.

• Productivity & Workflow was the largest cross-cutting learning topic, appearing in 50 sessions, including 38 AI-focused sessions.

• Workflow / productivity was the broadest theme, appearing in 56 sessions.

• 39 research abstracts showed that the SIIM AI pipeline remains strongly anchored in academic and health-system settings.

• The speaker mix suggests SIIM is less about AI hype and more about implementation, governance, informatics, and deployment inside imaging environments.

Key takeaway: After examining AI sessions, exhibitors, and product launches around ECR, the SIIM 2026 landscape shows the next layer of the same shift: radiology AI is moving from standalone model performance into workflow integration, enterprise imaging, governance, and deployment infrastructure.

🧲 Prostate MRI AI more than doubled PSA specificity

RadAISlice: A prostate MRI model points toward fewer unnecessary biopsies, if external and prospective performance holds.

The details:

• Trained and validated on 4401 patients across 6 MRI cohorts.

• Validation AUROC ranged from 0.875 to 0.966.

• Expert reader study showed AUROC 0.907 vs 0.805.

• With PSA, specificity rose from 15% to 38%.

Key takeaway: This could sharpen biopsy triage in prostate MRI, but I would want prospective workflow testing before relying on it for rule-out decisions.

🧠 ICH AI shows the cost of false positives

RadAISlice: This survey gives a grounded reminder that AI adoption depends on workflow, not just clearance or model performance.

The details:

• 65 teleradiologists used an FDA-cleared ICH overlay.

• Only 18.5% found false positive alerts acceptable.

• 10.8% reported reduced interpretation time.

• 33.8% said false positive review outweighed benefits.

• Trust fell to 3.1% when AI conflicted with reads.

Key takeaway: This fits the SIIM signal: radiology AI is becoming an implementation problem. Specificity, latency, alert design, and workflow fit can decide whether radiologists actually use the tool.

🩺 Breast ultrasound AI boosted reader accuracy

RadAISlice: This multicenter ultrasound study stands out for external testing and reader impact.

The details:

• 3048 women and 24762 images came from 5 hospitals.

• External cohort included 649 women and 5422 images.

• BrcaDetect combined images, BI-RADS, and demographics.

• Reader accuracy rose from 0.919 to 0.977 with AI.

Key takeaway: For breast ultrasound, the practical value is not just AUC. It is whether AI reduces false positives while preserving sensitivity.

Hepatic Vessel Map (HVM) (December 9, 2025)

Modality: CT | Focus: liver; hepatic vessels | Task: vessel segmentation; liver tumor segmentation

• Size: 282 CT scans from 282 patients. Dual-center, over 41,400 slices.

• Annotations: Full-volume masks for hepatic veins, portal veins to 3rd-order branches, and liver tumors. Expert-reviewed.

• Institutions: The University of Hong Kong-Shenzhen Hospital; Peking University Shenzhen Hospital

• Availability:

  public: Zenodo

• Highlight: Separate fine-grained hepatic and portal vein labels. Includes many diseased livers and preoperative planning cases.

🏛️ FDA Clearances

• K260324 - Canon cleared AI based CT software to support TAVR planning from cardiac and vascular anatomy.

• K260234 - Smart Alfa cleared AI software for automated processing of musculoskeletal radiology images.

• K254131 - DeepHealth cleared BAC software for automated radiologic image processing.

• K253831 - Cercare cleared software to assess capillary function on MRI, CT, and CBCT images.

• K260113 - United Imaging cleared an MRI system for diagnostic imaging and treatment planning.

• K253720 - United Imaging cleared pulsed Doppler ultrasound systems for radiology imaging.

• Explore last week's 10 radiology AI FDA approvals.

📄 Fresh Papers

• doi:10.1148/ryai.250331 - Externally tested CT nodule AI showed 88% sensitivity, 75% specificity, and AUROC 0.89 across 7454 nodules.

• doi:10.3174/ajnr.A9445 - A CT perfusion model for acute lacunar stroke reached AUC 0.82, outperforming neurologist consensus at AUC 0.58.

• doi:10.1055/a-2857-0974 - Deep learning MRI reconstruction cut exam time 11%, raised throughput 7.2%, and reduced repeats 25% in 8183 exams.

• doi:10.2196/87368 - A multipass LLM report QA framework cut human review burden from 192 to 88 reports per 1000 while preserving detection.

• Browse 189 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• A retrospective French screening study estimated 77% lower initial mammography reading workload with AI first read triage.

• Subtle Medical raised $33 million to expand AI tools for faster MRI and PET workflows across about 1300 scanners.

• Mayo researchers used routine pathology slides from 672 patients to predict meningioma class and recurrence risk.

• A clinician and patient survey flagged bias, hallucinations, accountability, and validation as key AI trust concerns.

• Read 10 fresh radiology AI news articles for last week.

That's it for today!

👋 Quick favor: drag this into your Primary tab so you don’t miss next week. Or just hit Reply with one thought.

See you next week.

P.S. We keep building free tools to accelerate your radiology work. What's the most time-consuming pain point in your day that we should help speed up? Reply and share your take so we keep building around you.