Chest CT AI cut reporting time 14.6%

Good morning, there. Commercial nodule AI cut median chest CT reporting time by 14.6%.

I see this as one of the clearer real world workflow signals for imaging AI. The effect was modest per case, but at 20000 plus annual CTs it translated into about 0.5 FTE.

Where would this kind of time saving matter most in your CT practice?

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

• Chest CT nodule AI shows real workflow time saved

• Mammography AI gets autonomous triage approval

• ACR tests LLM monitoring for imaging AI

• DL cine MRI cuts pediatric scan time

• Plus: 2 newly released datasets & 4 new papers.

BUILT BY US

GPT-6 Astra is here! 🚀

🫁 Chest CT nodule AI shows real workflow time saved

RadAISlice: This Radiology study gives a practical look at AI time savings in chest CT.

The details:

• 39323 chest CT exams across 19433 patients

• Median reporting time fell from 21.3 to 18.2 minutes

• Adjusted report completion improved with HR 1.17

• Estimated annual workload reduction was about 0.5 FTE

Key takeaway: This shifts the AI value discussion from detection alone to measurable radiologist capacity in high volume chest CT workflows.

🦩 Mammography AI gets autonomous triage approval

Image from: Radiology Business

RadAISlice: This approval pushes breast AI from assistive reads toward supervised autonomy.

The details:

• Vara received CE Certification under EU MDR

• AI can mark normal screening mammograms without radiologist review

• Abnormal or uncertain exams still go to at least one radiologist

• Atmon monitoring can revert cases to full review if quality falls

Key takeaway: For screening programs, this may become an early template for autonomous AI with monitoring and human oversight built into deployment.

📊 ACR tests LLM monitoring for imaging AI

RadAISlice: This registry work feels important because adoption needs surveillance, not just clearance.

The details:

• Assess AI prompts cover 9 monitored use cases

• Final prompt agreement was 0.985 for ICH

• Final prompt agreement was 0.997 for PE

• Independent accuracy and utility remain unestablished

Key takeaway: This points toward continuous performance monitoring as core infrastructure for safe imaging AI deployment.

❤️ DL cine MRI cuts pediatric scan time

RadAISlice: This prospective CMR study shows acceleration can fit real pediatric workflows.

The details:

• 50 pediatric patients, 56% with congenital heart disease

• Short axis cine time fell 66%, 83%, and 89%

• Diagnostic image quality held in 96%, 88%, and 88%

• LVEF and RVEF differed by 2 percentage points or less

Key takeaway: For pediatric and congenital CMR, DL reconstruction may reduce breath hold burden while keeping ventricular quantification usable.

2024 MRSI Challenge Synthetic Dataset (2026-09-02)

Modality: MRSI | Focus: Brain | Task: Nuisance-signal removal; metabolite quantification

• Size: 32 synthetic 3D FID-MRSI datasets from 32 HCP subjects. 24 train and 8 test.

• Annotations: Ground-truth metabolite maps and signals. Water+lipid nuisance labels; MM and baseline for test ground truth.

• Institutions: Technical University of Munich; Eindhoven University of Technology; et al.

• Availability:

  Public: Zenodo

• Highlight: Controlled 3T brain FID-MRSI benchmark with realistic B0 effects, water/lipid leakage, and known metabolite ground truth.

MYOSAIQ (2026-08-28)

Modality: MRI | Focus: heart, myocardium | Task: segmentation, infarct quantification

• Size: 439 LGE CMR exams from 439 patient time points. D8 123, M1 187, M12 129.

• Annotations: Manual 3D masks for LV cavity, healthy myocardium, MI, and MVO. Test set has second-expert labels.

• Institutions: CREATIS, Hospices Civils de Lyon, et al.

• Availability:

  request-only: Human Heart Project

• Highlight: Heterogeneous LGE CMR dataset across 16 centers, 3 MRI vendors, and acute to chronic post-MI phases.

📄 Fresh Papers

• doi:10.2196/92181 - A prospective study of 1054 chest radiographs found AI triage and report generation cut TAT by 90.6%.

• doi:10.1148/ryai.260100 - FDA authors warned that AI reconstruction can improve apparent image quality while introducing new failure modes.

• doi:10.1016/j.lungcan.2026.109596 - Commercial CT AI detected more pathologically sampled thoracic lymph node stations than routine reports documented.

• doi:10.1007/s00330-026-12830-y - A CT topogram framework labeled anatomy across 65250 patients with external F1 up to 0.970 for classification.

• Browse 233 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• Elucid raised $55M to expand AI cardiovascular CT analysis and support FFR CT development under FDA review.

• BioPINN LM estimated CTA derived aortic wall stress in under 3 seconds, but still needs outcome validation.

• A whole body MRI AI pipeline linked body composition changes with survival signals in multiple myeloma.

• AI analyzed resident case exposure to identify pathology gaps and guide personalized radiology teaching cases.

• A chest X ray false positive case underscores why AI alerts still need clinical context and radiologist oversight.

• Read 13 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.