MRI AI predicts outcomes in non-ischemic cardiomyopathy (ECV HR 3.09)

RadAI Slice Newsletter

Weekly Updates in Radiology AI

Good morning, there. AI-automated CMR quantification of ECV predicts outcomes in NIDCM (HR 3.09 for ECV ≥30%).

I see this as a significant milestone—the integration of automated CMR tissue mapping with AI streamlines risk stratification in non-ischemic dilated cardiomyopathy (NIDCM), offering reliable predictions for adverse outcomes and reverse remodeling. This finding feels especially relevant as it brings precision quantification into routine cardiac MR workflows and could prompt wider adoption among radiologists focused on heart failure care.

Would automated tissue mapping shift your management of NIDCM patients?

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

• AI-automated CMR mapping improves risk stratification in NIDCM

• FDA clears 56 new AI radiology tools, approvals top 1,000

• Multicenter MRI AI predicts rectal cancer response and immune microenvironment

• FDA-cleared imaging AI proliferates but reimbursement remains rare

• Plus: 3 newly released datasets, 6 FDA approved devices & 4 new papers.

🧬 AI-automated CMR mapping improves risk stratification in NIDCM

RadAI Slice: A new study shows that AI-automated CMR mapping for ECV predicts both risk and recovery in NIDCM.

The details:

• 347 NIDCM patients from two centers included (median follow-up: 38 months)

• Automated ECV ≥30% predicted adverse events (HR: 3.09, 95% CI: 1.87-5.12, p<0.001)

• Higher ECV by AI mapped to less functional improvement (reverse remodeling)

• Integration into workflow may enhance precision and efficiency

Key takeaway: This study supports automated CMR tissue mapping as a practical, data-driven risk stratification tool in NIDCM, potentially guiding heart failure management and radiology reporting.

🚀 FDA clears 56 new AI radiology tools, approvals top 1,000

Image from: Radiology Business

RadAI Slice: Radiology now accounts for over 1,000 FDA-cleared AI devices, underscoring rapid sector growth.

The details:

• FDA's list surpasses 1,300 AI-enabled devices overall (1,039 are radiology tools)

• 56 AI radiology products cleared from July-September 2024

• Big names: GE HealthCare, Siemens, Fujifilm, Qure.ai among latest clearances

• 80% of all FDA-cleared medical AI devices now target radiology

Key takeaway: The sheer volume of FDA-cleared radiology AI demonstrates innovation—yet underscores the need for evidence and reimbursement to translate these tools into routine clinical impact.

📺 Multicenter MRI AI predicts rectal cancer response and immune microenvironment

RadAI Slice: A transformer-based MRI AI model stratifies response and immune landscape in 1,026 LARC patients.

The details:

• Multicenter cohort: 1,026 rectal cancer patients; robust external validation

• MR-DELTAnet AUC: 0.93/0.88/0.90 in train/test/validation for pCR prediction

• Immune microenvironment linked with MRI AI-derived risk scores (single-cell RNA-seq)

• Model identifies distinct survival (DFS/OS) differences by prediction

Key takeaway: High generalizability and biological correlation make this an important step for individualized MRI-guided rectal cancer care and multi-modal integration in radiology oncology.

📄 FDA-cleared imaging AI proliferates but reimbursement remains rare

Image from: Radiology Business

RadAI Slice: Despite hundreds of cleared imaging AI tools, very few receive reimbursement.

The details:

• 80% of all FDA-cleared AI is for radiology (hundreds of tools)

• Only 2 Medicare Category I CPT codes (both in cardiac AI; 2nd arrives in 2026)

• Most algorithms lack demonstration of patient benefit in outcomes

• Reimbursement gap slows adoption despite regulatory progress

Key takeaway: Bridging the policy gap between FDA clearance and reimbursement will be essential to move AI from promising pilot to standard radiology practice.

Walnut-PCCT (2025-11)

Modality: CT | Focus: Non-human head, bones | Task: Spectral CT reconstruction, material decomposition

• Size: 15 walnuts, 172,800 projections, 15 scans × 4 beds × 1440 views × 2 energies

• Annotations: Raw projections with system parameters, material decomposition maps, virtual monoenergetic reconstructions

• Institutions: Huazhong University of Science and Technology, Hainan University

• Availability:

  Public (Zenodo, see article for other sub-links)

• Highlight: First large-scale, real cone-beam photon-counting CT dataset with raw multi-energy projections and full calibration for spectral imaging research

MK-11 (2025-12-10)

Modality: Microscopy | Focus: Bone marrow, Hematopathology | Task: Classification, Subtype recognition

• Size: 7,204 images from 70 patients

• Annotations: 11 megakaryocyte subtype labels, expert-annotated with consensus and quality metrics

• Institutions: Xinjiang University, Xiangya Hospital, et al.

• Availability:

  Public (Figshare link)

• Highlight: First public dataset with detailed megakaryocyte subtype labels and patient-level splits for robust, fair AI benchmarking.

MCT-LTDiag (2025-11-19)

Modality: CT | Focus: Liver | Task: Classification, Segmentation

• Size: 517 patients (2068 four-phase scans)

• Annotations: Expert-validated tumor masks, liver masks

• Institutions: Peking Union Medical College Hospital, Fudan University et al.

• Availability:

  Public (Harvard Dataverse)

• Highlight: First public multi-phase CT dataset for 5 major liver tumor subtypes with rigorous expert annotation and standardized pipeline.

🏛️ FDA Clearances

• K251203 - AVIEW Lung Nodule CAD automates lung nodule detection on CT, supporting early intervention in lung cancer screening.

• K251514 - Overjet CBCT Assist provides AI-powered dental/maxillofacial analysis in cone-beam CT, enhancing diagnostic accuracy and workflow efficiency.

• K250791 - ASUS Ultrasound Imaging System (LU800 series) delivers advanced pulsed Doppler and detailed diagnostic imaging to aid clinical assessment.

• K251601 - Mindray Hepatus/Fibrous series offers pulsed Doppler ultrasound for real-time imaging of tissue and vessels to inform diagnosis.

• K252498 - Sonoscape's Digital Color Doppler Ultrasound System supports real-time vascular imaging across a versatile hardware lineup.

• K251370 - Cartesion Prime (PCD-1000A/3) V10.21 (Canon Medical) delivers advanced PET imaging for various conditions using tomographic body scans.

• Explore last week's 7 radiology AI FDA approvals.

📄 Fresh Papers

• doi:10.1007/s00256-025-05106-x - AI model aids detection of TFCC injuries on wrist MRI, showing non-inferiority to radiology residents in multicenter data.

• doi:10.1038/s41598-025-27934-6 - Hybrid explainable DRM-Net model achieves 99% AUC for breast cancer classification on ultrasound, using transfer learning and XAI class activation mapping.

• doi:10.1177/00220345251387713 - OralSAM enables automated segmentation and quantitative measurement of periodontal tissue on intraoral ultrasound videos, matching inter-expert reliability.

• doi:10.1212/WNL.0000000000214527 - Brain-predicted age difference (brain-PAD) on MRI associates with disability and cognitive impairment in MS patients, even when controlling for chronological age.

• Browse 152 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• Deep learning model detects coronary microvascular dysfunction from ECGs, validated with PET, showing AUCs up to 0.955 for LVEF prediction.

• RNACOREX offers explainable, open-source AI to reveal microRNA–mRNA regulatory networks across 13 cancers, with survival stratification and transparent molecular mapping.

• Machine learning models combining clinical and molecular data accurately predict colorectal cancer survival, indicating advances toward precision oncology in radiology-AI integration.

• Read 5 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.