AI MRI model boosts early liver cancer detection to 87.9% sensitivity

RadAISlice Newsletter

Weekly Updates in Radiology AI

Good morning, there. A multicenter MRI-based AI achieved 87.9% sensitivity for early perihilar cholangiocarcinoma.

I’m struck by the model’s impact—detecting nearly 90% of early-stage cancers, far above radiologist performance. This is especially relevant as early detection is critical for potentially curative intervention in primary sclerosing cholangitis. AI may be narrowing a key clinical gap where expert readers still miss cases.

Would you incorporate an AI model like this into routine liver MRI reporting?

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

• AI outperforms experts in early cholangiocarcinoma detection on MRI

• Streamlined FDA pathway for radiology AI updates proposed

• Chest CT enhancement AI reduces repeats, boosts confidence

• FDA clears Neosoma Brain Mets AI for brain metastasis detection

• Plus: 2 newly released datasets, 5 FDA approved devices & 4 new papers.

🧬 AI outperforms experts in early cholangiocarcinoma detection on MRI

RadAISlice: AI analysis of MRI scans outperformed senior radiologists in identifying early-stage liver cancer.

The details:

• 398-patient, multicenter, international cohort with 230 cancer cases analyzed

• Test set: AI sensitivity 87.9% vs radiologists 50%, specificity 84.1% vs 100%

• Model surpassed radiologists even when no visible mass present (91.6% vs 50.6%)

• Study focused on PSC patients being evaluated for liver transplantation

Key takeaway: This externally validated AI tool may raise early cholangiocarcinoma detection rates, improving eligibility for curative therapies and guiding surveillance.

⚖️ Streamlined FDA pathway for radiology AI updates proposed

RadAISlice: US regulators are evaluating public input on a less burdensome pathway for radiology AI innovation.

The details:

• Streamlined pathway would let some cleared AI bypass new 510(k) for feature updates

• Postmarket study requirements would replace premarket review in defined cases

• Targets CADx, CADe/x, CADt, and image analyzer AI already cleared by FDA

• FDA accepting comments until Feb 2026—a critical window for radiology stakeholders

Key takeaway: If adopted, this could speed US adoption of updated imaging AI solutions—affecting product rollouts and lifecycle management for radiology departments.

🩺 Chest CT enhancement AI reduces repeats, boosts confidence

RadAISlice: New AI algorithms markedly improve the quality of challenging chest CT and CTPA exams.

The details:

• 611 chest CT/CTPA studies analyzed, 318 labeled as suboptimal by thoracic radiologists

• AI increased main pulmonary artery SNR from 20.8 to 59.7, CNR from 188.8 to 288.8

• Pulmonary artery attenuation: 192 to 293 HU, ascending aorta: 235 to 362 HU (p=0.001)

• Presenters were from Massachusetts General Hospital and Harvard Medical School

Key takeaway: AI enhancement could improve diagnostic yield and workflow while reducing unnecessary repeat scans—a practical advance for busy departments.

🖥️ FDA clears Neosoma Brain Mets AI for brain metastasis detection

RadAISlice: An FDA-cleared AI now assists radiologists by automatically detecting brain metastases on MRI.

The details:

• Neosoma Brain Mets is an AI software for radiologists, cleared by FDA via 510(k)

• Performs automated detection—aims to improve efficiency and reproducibility

• Relevant for neuro-oncology and broader MRI brain cancer workflows

• First-wave adoption may be in cancer centers managing complex patients

Key takeaway: FDA-cleared AI for brain mets brings more automated, reproducible reporting into practice—particularly impactful for high-volume cancer imaging.

MBB-MRI (2025-11-14)

Modality: MRI | Focus: Whole body, Musculoskeletal | Task: Segmentation, Morphometry

• Size: 102 healthy adults; 70+ muscles and 13 bones segmented per subject

• Annotations: 3D segmentations of 70 muscles and 13 bones; labels include volume, fat fraction, asymmetry

• Institutions: Springbok Analytics, University of Virginia, Ghent University, et al.

• Availability:

  request-only (see paper)

• Highlight: Most comprehensive in vivo full-body muscle and bone MRI dataset with AI-assisted 3D segmentations.

P-ELCAP (2025-12-27)

Modality: CT | Focus: lung | Task: classification, risk prediction

• Size: 211 participants, 211 LDCT scans (67 cancer, 68 benign nodules, 71 control, 5 false positives)

• Annotations: Nodule-level segmentations (malignant, benign, false positive), clinical and demographic data, proteomics (>1000 plasma proteins)

• Institutions: Clínica Universidad de Navarra, Institute of Physics of Cantabria, et al.

• Availability:

  public / request confirmation for lung cancer research via Zenodo

• Highlight: First open multimodal LDCT dataset with plasma proteome (>1000 proteins) and detailed nodule labels from a lung screening cohort

🏛️ FDA Clearances

• K252670 - Alzevita, FDA-cleared image processing software, automates radiology workflow tasks to support interpretation.

• K252029 - AI-CVD is FDA-cleared to support automated cardiovascular imaging analysis, streamlining detection tasks.

• K250976 - Airbile-100 is a mobile X-ray unit, FDA-cleared, enabling flexible imaging across care settings.

• K252235 - PVAD IQ Software, an FDA-cleared product, automates steps in radiological image processing to improve efficiency.

• K251416 - UltraSight Guidance, cleared by FDA, features AI-powered tools to guide and optimize radiological image acquisition.

• Explore last week's 11 radiology AI FDA approvals.

📄 Fresh Papers

• doi:10.1136/bmjopen-2025-111360 - In systematic review/meta-analysis, AI in breast cancer screening matched or improved sensitivity and reduced workload as a second or triage reader.

• doi:10.1161/CIRCIMAGING.125.018353 - AI-enabled echocardiography provided accurate aortic stenosis risk stratification, correlating with CT and PET severity and predicting valve replacement.

• doi:10.1016/j.scib.2025.11.027 - DeepSTEMI, trained/validated on 944 patients, achieved AUC 0.89 for 2-year outcomes after STEMI, outperforming manual CMR parameters in clinical prediction.

• doi:10.1038/s41746-025-02260-3 - A 3D deep learning tool (PAN-VIQ) quantified vascular invasion in pancreatic cancer, matching senior radiologist accuracy in prospective validation.

• Browse 97 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• Deep learning image reconstruction improved image quality and dose reduction in dual-energy CT venography for liver transplant assessment.

• Machine learning integrates carotid ultrasound and brain MRI for cerebral blood flow prediction in spaceflight analogs.

• A multicenter deep learning approach predicts post-implantation language outcomes with 92% accuracy using pediatric brain MRI.

• LSTM-based algorithms triple range and retain sensitivity in optical interferometry, suggesting improvements for real-time medical monitoring.

• Read 6 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.