AI scores thymic health on CT predicts cancer and heart outcomes

Good morning, there. AI analysis of CT scans in over 25,000 adults linked higher thymic health scores to ~50% lower all-cause mortality.

I see this finding as a major step toward leveraging routine imaging for actionable, prognostic biomarkers. It supports the growing role of radiology in personalized medicine and may influence future patient stratification strategies. Such advanced AI-derived measures could impact oncologic and cardiovascular care far beyond current risk models.

How might you integrate AI-based biomarkers like thymic health into routine CT reads?

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

• AI Thymic Scoring on CT Tied to Survival and Immunotherapy Benefit

• AI Mammography Trial: Detection Up, Workload Down in Prospective Study

• Prospective RCT: AI Significantly Boosts Lung Nodule Detection

• Multicenter Deep Learning Yields Population Norms for Cardiac MRI

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

🧬 AI Thymic Scoring on CT Tied to Survival and Immunotherapy Benefit

Image from: EurekAlert

RadAISlice: Mass General Brigham researchers show AI-derived thymic health from CT strongly predicts survival and cancer response.

The details:

• Analyzed over 25,000 adults’ CTs for thymic health with deep learning models

• Higher health: ~50% lower all-cause mortality, 36% lower lung cancer risk

• In 1,200+ cancer patients, healthy thymus linked to 44% lower risk of death post-immunotherapy

• Chronic inflammation, BMI, smoking tied to lower thymic scores

• Study published in Nature; not yet ready for direct clinical use

Key takeaway: AI can derive meaningful prognostic markers from routine CT, and early adoption of such tools may expand the role of radiology in shaping patient management and risk stratification.

🩺 AI Mammography Trial: Detection Up, Workload Down in Prospective Study

RadAISlice: A large, prospective study found AI-based triage safely increased cancer detection and reduced radiologist workload by over 60%.

The details:

• Prospective trial, 31,301 women undergoing screening mammography

• AI excluded low-risk cases, reducing workload by 63.6% overall

• Cancer detection rate rose from 6.3 to 7.3 per 1,000; recall rate rose from 4.8% to 5.5%

• Positive predictive value preserved, despite screening more autonomously

• Workload dropped by 62-66% across DM and DBT modalities

Key takeaway: This is one of the strongest real-world demonstrations to date for safe partial AI triage in breast screening, potentially guiding future workflow design, regulation, and ethical questions about autonomous AI.

🫁 Prospective RCT: AI Significantly Boosts Lung Nodule Detection

Image from: Radiology Business

RadAISlice: A new prospective RCT showed that AI-assisted interpretation improved lung nodule detection in LDCT screenings.

The details:

• RCT compared radiologist-only to AI-assisted reads in consecutive LDCT screens

• Measured interpretation time, nodule detection, and follow-up recommendations

• AI significantly increased detection of actionable Lung-RADS-positive nodules

• First prospective real-world validation after many retrospective studies

Key takeaway: Robust, prospective evidence strengthens the case for integrating AI in lung cancer screening, supporting earlier detection and workflow efficiency in chest CT.

🤖 Multicenter Deep Learning Yields Population Norms for Cardiac MRI

RadAISlice: A population study used automated AI segmentation on nearly 30,000 CMR exams to define normal cardiac structure and function.

The details:

• 29,908 participants analyzed, main and healthy subcohorts defined

• Short-axis cine images, AI-based segmentation, manual QC

• Detailed reference percentiles for LVEF, RVEF, volumes by age/sex

• Supports standardized, AI-enabled reporting on large scale

Key takeaway: Radiologists now have AI-derived, population-level normative data for cardiac MRI, enabling sharper diagnostic thresholds and research into sex- and age-related remodeling.

Reaching-Multimodal Dataset (RMD) (2026-03-04)

Modality: US, MoCap, EMG | Focus: Upper arm, Muscles | Task: Point tracking, Biomechanics analysis

• Size: 36 participants, ~300,000 ultrasound frames

• Annotations: Frame-level tracking of 11 anatomical points in ultrasound; synchronized kinematics, EMG, accelerometry, tremor, event segmentations

• Institutions: MIT, Fraunhofer Portugal et al.

• Availability:

  Public (Figshare link)

• Highlight: First dataset with dense, frame-level ultrasound point tracking; supports deep learning and biomechanics research

DyABD (2026-01-29)

Modality: MRI | Focus: Abdomen, muscle | Task: Segmentation, quantification

• Size: 311 dynamic MRI volumes from 17 patients

• Annotations: Manual and semi-automatic segmentations of 4 abdominal muscle groups; expert verified

• Institutions: University College Dublin, Aix Marseille Univ et al.

• Availability:

  request-only (link)

• Highlight: First dynamic MRI dataset with pre- and post-operative abdominal muscle segmentations during exercises in hernia patients

ToothPix (2026-03-03)

Modality: X-ray | Focus: oral cavity | Task: segmentation, detection

• Size: 8,655 panoramic X-rays; 8,655 patients

• Annotations: 30,186 pixel-level lesion segmentations; polygonal contours for teeth and oral diseases, JSON and mask files

• Institutions: Changsha Stomatological Hospital, Central South University of Forestry and Technology, et al.

• Availability:

  restricted (Zenodo link)

• Highlight: High-res multi-lesion oral X-rays; largest with expert pixel-level annotations for segmentation and detection.

Ar-PlaqSegm1 (February 2026)

Modality: US | Focus: carotid artery | Task: segmentation

• Size: 541 images from unique patients

• Annotations: binary masks for atherosclerotic plaque, expert-validated

• Institutions: Universidad Nacional de Córdoba, Universidad Tecnológica del Uruguay

• Availability:

  public https://www.doi.org/10.17632/8srkpz52dy.1

• Highlight: First open B-mode carotid ultrasound segmentation dataset with plaques and plaque-free cases under heterogeneous imaging conditions

🏛️ FDA Clearances

• K260166 - Bunkerhill Contrast CAC: 510(k)-cleared CT system supporting detailed cross-sectional imaging for radiology.

• K260167 - Bunkerhill Contrast AVC: FDA-cleared CT imaging solution for diagnostic cross-sectional X-ray imaging.

• K260169 - AV Cardiac CT: Cardiac-focused CT scanner from Philips, 510(k) cleared for detailed heart imaging.

• K253593 - Clarius Ejection Fraction AI: FDA-cleared ultrasound software automates cardiac ejection fraction calculation.

• K253818 - Annalise Enterprise: AI radiology triage and auto-notification platform, 510(k) cleared for workflow support.

• K251901 - Magnifico Open: 510(k)-cleared open MRI platform for anatomic imaging, supporting various use-cases.

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

📄 Fresh Papers

• doi:10.1038/s41586-026-10243-x - A multi-institutional study shows deep learning-based thymic health from CT predicts immunotherapy outcomes and survival across several cancers.

• doi:10.1038/s41591-026-04277-x - A prospective paired trial: AI-based triage in mammography increased detection rate 15% and reduced read workload 63% vs standard double-reading.

• doi:10.64898/2026.03.14.26348373 - Multicenter Indian study externally validated AI (LIRA) for CXR triage, showing 84–99% sensitivity and 64–92% specificity for key pathologies.

• doi:10.64898/2026.03.04.26347460 - A multicenter external validation: DeepMS fuses lesion and normal-appearing white matter from MRI to diagnose multiple sclerosis, outperforming traditional criteria.

• Browse 141 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• A large US study found no significant improvement in breast cancer detection or recall rates using AI for single-reader mammography.

• AI increased detection of actionable lung nodules in LDCT but did not reduce interpretation time; follow-up recommendations increased.

• AI leveraging cardiac ultrasound and EHR predicted peak VO2 with 85% accuracy, aiding advanced heart failure diagnosis.

• Mass General Brigham developed AI to identify patients at risk of domestic abuse by mining imaging and clinical records.

• RSNA coverage shows vendors shifting from isolated AI to enterprise, workflow-integrated platforms for broader system value.

• Read 20 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.