CT AI cut ILD procedures from 38% to 16%

Good morning, there. Fibresolve was used in 209 patients across 47 centers, with planned invasive procedures dropping to 15.6%.

I read this as a shift from asking whether CT AI can classify ILD to whether it changes decisions. The signal matters because adoption depends on management impact, not AUC alone. It also gives radiologists a concrete outcome to scrutinize.

What outcome would make you trust CT AI in ILD workups?

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

• CT AI shifts ILD decisions in 47 centers

• Medicare pays for CT body triage AI

• Mammography AI lifts detection without slowing reads

• Breast ultrasound AI earns FDA clearance

• Plus: 5 newly released datasets, 6 FDA approved devices & 4 new papers.

🫁 CT AI shifts ILD decisions in 47 centers

RadAI Slice: This prospective CT AI study connects FDA authorized output to real care decisions.

The details:

• 209 patients across 47 US centers used Fibresolve in routine ILD care

• Positive CT AI result in 89 of 209 patients, or 42.6%

• ILD specific therapy rose from 12.5% to 44.4% after use

• Planned invasive procedures fell from 38.3% to 15.6% performed

Key takeaway: This moves imaging AI evidence from accuracy toward management impact, especially when CT can reduce diagnostic uncertainty in ILD.

💵 Medicare pays for CT body triage AI

Image from: Radiology Business

RadAI Slice: This reimbursement decision may make inpatient CT triage AI easier to justify operationally.

The details:

• NTAP begins October 1 for qualifying inpatient cases

• Payment is up to $137.53 per case for Aidoc CARE CT Body Triage

• CMS aims to offset about 65% of average device cost

• Hospitals must report ICD 10 PCS code XEZ5XKC

Key takeaway: This changes the adoption math for hospitals, linking radiology AI deployment to a defined Medicare payment pathway.

🎀 Mammography AI lifts detection without slowing reads

RadAI Slice: This prospective workflow study gives mammography AI a pragmatic test in daily practice.

The details:

• 4577 screening and diagnostic exams used alternating monthly AI display

• Cancer detection was 22.3 vs 11.5 per 1000 with AI

• Mean reading time was 65.0 vs 64.3 seconds

• Diagnostic AIR rose to 18.7% vs 12.1%, screening AIR did not

Key takeaway: This suggests workflow safe screening gains, while diagnostic mammography needs closer monitoring for added calls.

🔎 Breast ultrasound AI earns FDA clearance

RadAI Slice: This FDA clearance adds a structured breast ultrasound AI tool with reader study data.

The details:

• SMART B localizes suspicious lesions on two orthogonal views

• MRMC study included 589 patients and 16 radiologists

• Reader LROC AUC improved from 0.824 to 0.897 with AI

• Standalone sensitivity was 94.7%, specificity 51.4%

Key takeaway: This may help standardize breast ultrasound reporting, but specificity shows radiologist oversight remains central.

Time-Aware Multi-View MRI Benchmark (13 Aug 2026)

Modality: MRI | Focus: Brain, CNS tumors | Task: Temporal reasoning, disease progression localization

• Size: Over 3,200 MRI timepoints from 890 patients. 3–4 serial timepoints per patient.

• Annotations: 3,920 expert-verified QA pairs. Open-ended, MCQ, and binary labels with reasoning steps and localization guidance.

• Institutions: Mohamed bin Zayed University of Artificial Intelligence, Department of Health Abu Dhabi, et al.

• Availability:

  Public: GitHub

• Highlight: Unifies longitudinal, multi-view MRI with temporal change localization across seven public cohorts.

PatchChestCT (12 August 2026)

Modality: CT | Focus: Chest; cardiovascular/mediastinum | Task: Localization; classification

• Size: 2,201 non-contrast chest CT volumes/studies from 2,201 patients; 2,016 train and 185 validation.

• Annotations: Physician-reviewed 3D patch-level binary labels on a 12×12×24 grid for 9 abnormalities.

• Institutions: USTC; The First Affiliated Hospital of USTC, et al.

• Availability:

  Public annotations: Zenodo. Source CTs require CT-RATE DUA: CT-RATE.

• Highlight: Token-aligned 3D patch labels for scalable multi-abnormality localization in chest CT.

CT-∆Bench (12 Aug 2026)

Modality: CT | Focus: Thorax, lungs | Task: Longitudinal report generation, temporal change detection

• Size: 5,614 CT scans from 2,807 patients. 2,638 train pairs and 169 validation pairs.

• Annotations: LLM-synthesized difference reports. Atomic change events labeled NEW, RESOLVED, INCREASED, DECREASED, or STABLE.

• Institutions: University of Tennessee at Chattanooga; Chongqing Medical University; et al.

• Availability:

  Public: Hugging Face

• Highlight: Dedicated paired 3D CT benchmark for longitudinal difference reporting with patient-level splits and change-aware metrics.

RCC-AID (August 5, 2026)

Modality: CT | Focus: Kidney; RCC | Task: Segmentation; classification

• Size: 129 contrast-enhanced CT scans from 91 patients. Includes 85 clear cell, 26 papillary, and 18 chromophobe RCC scans.

• Annotations: Voxel-level masks for kidney, primary solid tumor, and cysts. Metadata includes subtype and contrast phase.

• Institutions: Radboudumc; Charité - Universitätsmedizin Berlin; et al.

• Availability:

  Public: Zenodo

• Highlight: Open TCGA-RCC CT annotations across clear cell, papillary, and chromophobe subtypes.

BreastMammo & DenseMammo (2026-08-10)

Modality: X-ray mammography / FFDM | Focus: Breast, breast density | Task: Breast density classification, pathology diagnosis

• Size: 3,374 images from 1,067 patients. BreastMammo: 894 images from 447 patients. DenseMammo: 2,480 images from 620 patients.

• Annotations: Expert ACR breast density labels. BreastMammo also has pathology-confirmed benign/malignant labels and BI-RADS scores.

• Institutions: Northwestern University, Istanbul University, et al.

• Availability:

  Public: BreastMammo and DenseMammo

• Highlight: Two new multi-view FFDM benchmarks for domain generalization across mammography vendors and sites.

🏛️ FDA Clearances

• K260852 - Siemens cleared MAGNETOM Flow MRI systems with Deep Resolve and AutoAlign tools for image quality and workflow.

• K261427 - United Imaging cleared uMI Panvivo PET CT with AI reconstruction, artifact reduction, and motion correction features.

• K261210 - Cleerly Plaque cleared as a CCTA module for automated coronary plaque identification and quantification.

• K253835 - Hermes Affinity cleared multimodality viewing and analysis with deep learning organ segmentation tools.

• K261835 - GE LOGIQ Focus cleared as a general ultrasound system with AI workflow tools for anatomy and preset selection.

• K260936 - RadioViewAI cleared as a cloud PACS viewer that can display outputs from FDA cleared third party AI algorithms.

• Explore last week's 18 radiology AI FDA approvals.

📄 Fresh Papers

• doi:10.3174/ajnr.A9575 - Multicenter CTA validation of Methinks LVO AI reported 95.8% sensitivity, 88.6% specificity, and 3.3 minute alerts.

• doi:10.1016/j.chest.2026.07.5237 - HANSE lung screening data showed CT airway biomarkers plus clinical factors improved COPD detection to AUC 0.83.

• doi:10.1016/j.jacr.2026.08.007 - A JACR DXA reporting pipeline cut academic report creation time from 3.48 to 0.87 minutes with 99.9% accuracy.

• doi:10.1148/radiol.260264 - Radiology published a 1181 patient CT radiomics model for lung adenocarcinoma high grade patterns and prognosis.

• Browse 318 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• Twelve systems caring for nearly 20 million patients formed a consortium to measure AI triage and workflow impact.

• ABR says it is not using AI for certification decisions and is forming advisory groups around ethics and oversight.

• NYU reported 313335 DBT exams for 5 year risk prediction, with AUC 0.72 and calls for external validation.

• A thoracic imaging study found expertise reduced acceptance of incorrect LLM advice, with OR 0.54.

• An SBI survey found 47% had breast AI implemented, but only 34.7% reported lower recall rates.

• Read 18 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.