Trial finds under-reliance on correct AI hurts head CT reads; Penumbra thrombectomy cleared for DVT; Coreline announces FDA clearance

• AI Oct 23, 2026
  Under-reliance on correct AI advice hurts head CT reads more than over-reliance in Lancet Digital Health crossover trial
  In a randomized crossover experiment published Oct. 23 in The Lancet Digital Health, 50 radiologists and 75 non-radiologist physicians and medical trainees read 50 head CT scans for intracranial hemorrhage with no advice, basic AI advice or explainable AI advice. The advice was correct in 80% of cases, and under-reliance on it was more detrimental to diagnostic performance than over-reliance, the authors reported. Correct explainable advice improved performance (odds ratio 1.84, 95% CI 1.49 to 2.28, P < .0001) but did not reduce reading time versus no advice.
  Sources: The Lancet Digital Health, PubMed
• FDA Oct 20, 2026
  FDA clears Penumbra's Lightning computer assisted vacuum thrombectomy for deep vein thrombosis
  The Food and Drug Administration (FDA) cleared Penumbra's Lightning computer assisted vacuum thrombectomy (CAVT) technology for deep vein thrombosis (DVT), the company said Oct. 20. Dr. Patrick Muck, a TriHealth vascular surgeon and BOLT-DVT co-principal investigator, said the trial data produced the first cleared DVT indication among mechanical thrombectomy devices. In the company-reported trial, a prospective, multicenter investigational device exemption study, 98% of patients had complete or near-complete thrombus removal, with a median device time of 18 minutes. The 48-hour major adverse event rate was 3.1%, and the re-thrombosis rate within 30 days was 3.9%, the company said.
  Sources: Penumbra, Endovascular Today
• AI Oct 23, 2026
  Coreline Soft announces FDA clearance of AVIEW Aorta Metrics, AI software that measures aortic diameters on chest CT
  Coreline Soft said Oct. 23 that it received Food and Drug Administration (FDA) 510(k) clearance for AVIEW Aorta Metrics, software designed to automatically segment the aorta from chest CT images, classify aortic zones and measure the diameter of each zone. The clearance, received Oct. 16 and announced Oct. 23, is the company's 13th FDA 510(k) clearance, according to the company, which described the product as AI-powered.
  Source: Coreline Soft (PR Newswire)
• AI Oct 22, 2026
  AI assistance does not offset night-shift fatigue in residents reading bedside chest x-rays, European Radiology study finds
  In a study published Oct. 22 in European Radiology, 10 radiology residents each read three sets of 33 bedside chest x-rays before and after a first-call night shift, with and without AI assistance. Six of the 10 were classified as fatigued, and their accuracy fell from a kappa of 0.71 to 0.61 (P < .001), versus 0.72 to 0.65 (P = .037) in non-fatigued readers, while reporting times stayed stable. AI assistance did not significantly improve accuracy in either group and significantly shortened reporting time only in non-fatigued readers, by 9.9 seconds, the authors reported.
  Sources: European Radiology, Springer, AuntMinnie
• AI Oct 21, 2026
  Agentic AI system assigns BT-RADS scores to glioma follow-up MRI with 76.0% accuracy versus 57.5% for initial assessments
  A multiagent system combining a large language model and a convolutional neural network automated BT-RADS classification of posttreatment glioma MRI, according to a retrospective single-center study published Oct. 21 in Radiology: Artificial Intelligence. Of 509 consecutive examinations, 492 were evaluable; against expert annotation as the reference standard, the system's accuracy was 76.0% (374 of 492) versus 57.5% (283 of 492) for initial assessments (P < .001), and its Cohen kappa was 0.71. The authors said the results support further evaluation as a clinical decision-support tool.
  Sources: Radiology: Artificial Intelligence, PubMed
• Research Oct 20, 2026
  Photon-counting CT gives better image quality at lower dose than conventional CT in 51 patients with obesity
  In a retrospective intraindividual comparison published Oct. 20 in Radiology, 51 patients with obesity underwent both photon-counting CT and energy-integrating detector CT. Median CT dose index volume was 11.99 versus 17.05 mGy and median dose length product 526.25 versus 709.32 mGy·cm with photon-counting CT, and mean signal-to-noise ratio in the ascending aorta was 19.65 versus 13.22. The authors concluded that photon-counting CT had superior quantitative and subjective image quality at a lower radiation dose.
  Sources: Radiology, PubMed, AuntMinnie

The weekly innovation roundup for radiology, published every Sunday for the week before; every item links to its sources. Latest radiology news · All specialties