Weekly AI digest

Three things that mattered

• Renal cancer dominates with four fusion models

  Four of the eight selected papers target renal disease, all fusing imaging with pathology or clinical data and reading it out through SHAP. They range from early renal fibrosis on multimodal ultrasound (AUC 0.948 internally) to a YOLOv11 CT subtyping system running at 0.24 seconds per file.

• AI quantifies free air on CT

  A deep learning model for pneumoperitoneum on abdominal CT across 2072 patients reached AUC 0.97 in testing and 84.3 percent sensitivity in a 214-scan external emergency cohort. Its novel contribution is a reproducible free-air volume (ICC 0.996), which turns a binary alert into a potential triage threshold.

• Explainability becomes a baseline expectation

  Seven of the eight peer-reviewed selections are built around an interpretability layer, either SHAP or Grad-CAM. Explainability has shifted from a selling point to a baseline expectation, raising reproducibility while narrowing methodological diversity across a notably homogeneous, retrospective, China-based pool.

Peer-reviewed

Selected from 315 papers indexed in PubMed this week.

• Explainable Machine-learning Model Based on Multimodal Ultrasound for Non-invasive Detection of Early Renal Fibrosis: A Multicenter Study

  Academic Radiology, 14 July 2026

• An Explainable Multimodal 2.5D Deep Learning-Radiomics Model for Predicting Extranodal Extension in Lung Adenocarcinoma Using Preoperative CT: A Multicenter Retrospective Cohort Study

  BMC Medical Imaging, 15 July 2026

• A Multimodal, Multitask Prediction Framework for Diagnosis and Prognosis of Clear Cell Renal Cell Carcinoma

  Journal of Imaging Informatics in Medicine, 13 July 2026

• Deep Learning-Based CD8+ T Cell Model for Predicting Prognosis and Targeted Immunotherapy Benefit in ccRCC

  npj Digital Medicine, 13 July 2026

• Clinical Study on the Prevention of High-Risk Pulmonary Nodule Progression With Yifei Sanjie Pill: Protocol for a Multicenter Randomized Controlled Trial

  JMIR Research Protocols, 15 July 2026

• Deep Learning-Based Segmentation and Subtype Prediction of Renal Cell Carcinoma on Contrast-Enhanced CT

  npj Precision Oncology, 14 July 2026

• Diagnostic Performance of an Artificial Intelligence Algorithm for Detecting Pneumoperitoneum on Abdominal CT Scans

  Insights into Imaging, 18 July 2026

• A CT-Based Deep Learning Model for the Automated Risk Stratification of Refractory Mycoplasma Pneumoniae Pneumonia in Children

  BMC Medical Imaging, 17 July 2026

Industry & regulation

• Provect AI wins FDA clearance for CT-free 3D imaging

  Provect AI received FDA 510(k) clearance for 3D-Anywhere software that reconstructs volumetric 3D images from standard C-arm X-rays for spine and orthopedic procedures, alongside a 7 million dollar raise.

• Topcon secures CE mark for GAIA retinal imaging

  Topcon Healthcare received CE mark approval for GAIA, a fully automated ultra-widefield retinal imaging device offering a 209 degree field of view in a single capture.

• Raidium launches R.Read in the US

  Raidium launched its AI-native oncology imaging viewer R.Read in the United States, already deployed at Moffitt Cancer Center where it replaced a legacy radiomics tool.

• Nucs AI and Segmed partner on biomarker validation

  Nucs AI and Segmed announced a strategic partnership, including a Segmed investment, giving Nucs AI access to more than 2800 sites and 150 million imaging studies to validate predictive imaging biomarkers.

• GE HealthCare and Catholic Health form 500 million dollar alliance

  GE HealthCare and Catholic Health announced a 10-year, roughly 500 million dollar alliance covering AI-enabled MR, CT and PET systems and cloud radiology operations tools across more than 40 sites.

• Deepnoid commercializes generative AI M4CXR

  Deepnoid began commercial sales of M4CXR, a generative AI chest X-ray device that drafts a preliminary report across more than 41 finding types, after regulatory approval in South Korea.

Conferences & voices

RSNA News highlighted a UCSF study in Radiology Advances showing that AI-generated summaries of lung cancer screening reports improved comprehension and reduced anxiety across 1815 respondents. The week also brought two cautionary journal signals amplified in the trade press: a Radiology study on automation bias, where median sensitivity on false-negative AI suggestions fell to 39 percent with AI versus 71 percent without, and a Clinical Radiology study finding radiologists correctly identified AI-generated images only 75.0 percent of the time.

From the congress circuit, results from the multicenter AI INFORM study presented at SCCT showed that notifying physicians of AI-detected coronary plaque on chest CT increased lipid-lowering therapy by roughly 33 and 46 percent at 6 and 12 months, with a corresponding drop in LDL-C.

Among key voices, Curt Langlotz used LinkedIn to amplify the reporting and ambient-documentation layer as radiology AI's commercial battleground, resharing Bunkerhill Health's 55 million dollar raise and Cognita's ambient reporting pitch. Woojin Kim, in his ACR Data Science Institute role, reshared a post asking how to define an AI error, underscoring that a shared error definition is the prerequisite for meaningful post-deployment AI quality monitoring.

Source articles are indexed in PubMed with verified DOIs. Manuscript-stage work is excluded. The full weekly report is produced in French.

Get the weekly digest

One email a week on what matters in radiology and imaging AI. Researched by a radiologist and imaging informatics professional. Free, no vendor sponsorship.

By subscribing you agree to receive the weekly digest by email. Unsubscribe anytime. No sharing with third parties.