This Week in Radiology — Jun 16, 2026

Welcome to This Week in Radiology. This week we're covering 10 notable papers spanning the evolving role of artificial intelligence in diagnostics, advancements in imaging protocols that enhance safety and efficiency, and new insights that refine our diagnostic pathways. Let's dive in.

We begin with a look at artificial intelligence, where a trio of papers provides a crucial snapshot of its current capabilities and limitations. A study in Nature Medicine evaluated the performance of AI tools in medicine, comparing general-purpose large language models like GPT-5.2 and Gemini 3.1 Pro against specialized clinical AI tools such as OpenEvidence and UpToDate Expert AI [1]. Across three different benchmarks—testing medical knowledge, alignment with clinicians, and responses to real clinical queries from physicians—the general-purpose models consistently outperformed their specialized counterparts. In fact, the clinical AI tools performed similarly to a standard Google Search with its AI Overview feature. This finding underscores the urgent need for independent, real-world evaluation before these tools are widely integrated into clinical practice. Two papers from the American Journal of Neuroradiology provide more specific context on AI in image interpretation. One study retrospectively evaluated an AI tool designed to detect acute cervical spine fractures on CT scans from over 2600 exams [7]. While the AI demonstrated a high specificity of over 99%, its sensitivity was 89.1%, correctly identifying 98 of 110 acute fractures. This was statistically lower than the performance of radiologists, who achieved a sensitivity of 94.5%. The AI's false positives were often due to mimics like atherosclerotic calcifications, chronic fractures, and osteophytes. In a second AJNR paper, researchers assessed the performance of multimodal large language models, including newer versions like GPT-5 and Gemini 2.5, on neuroradiology multiple-choice questions [10]. While the latest models showed clear generational improvement and even approximated the performance of radiology residents, they still fell substantially short of expert neuroradiologists, who achieved a mean accuracy of over 91%. The study highlights that while aggregate performance is improving, a significant gap to the expert reference remains, and consistency on individual decisions is not guaranteed. Together, these papers suggest that while AI is advancing rapidly, human expertise remains the benchmark, and rigorous validation against expert performance is essential.

Next, we turn to several studies focused on optimizing imaging protocols to improve diagnostic accuracy, safety, and workflow. From the European Journal of Radiology, a prospective study investigated the feasibility of low-dose carotid CT angiography for quantitative plaque evaluation [9]. In a study of 67 patients who received both low-dose and conventional-dose scans, the low-dose protocols, using as low as 60 kVp, reduced the effective radiation dose by approximately 66%. Critically, this significant dose reduction did not compromise the accuracy of quantitative plaque analysis, with no significant differences in total plaque volume, component volumes, or stenosis measurements compared to the conventional protocol. Staying with CT, another paper in the European Journal of Radiology explored the value of dual-energy CT for assessing therapeutic response in hepatocellular carcinoma after transarterial chemoembolization, or TACE [6]. Comparing 40-keV monoenergetic images to conventional CT reconstructions in 48 patients, the study found that dual-energy CT significantly improved the detection of viable residual tumors, particularly for lesions smaller than 2 cm. It also enhanced tumor boundary delineation and improved interobserver agreement, with the greatest benefit seen among less-experienced readers. Shifting to MRI, a prospective study also in the European Journal of Radiology evaluated an ultra-rapid, unsedated, non-contrast MRI protocol for suspected pediatric appendicitis in 73 children [3]. With a mean acquisition time of just over 9 minutes, the protocol achieved a sensitivity of nearly 96% and a specificity of 96%, with an almost perfect interobserver agreement. The negative appendectomy rate was very low at 4.3%, and importantly, MRI identified alternative diagnoses in nearly 39% of children who did not have appendicitis. This positions rapid MRI as a powerful radiation-free and sedation-free option for this common pediatric emergency. Finally, in breast imaging, a multicenter prospective study in European Radiology compared automated breast ultrasound, or ABUS, with traditional handheld ultrasound as an adjunct to mammography for preoperative staging of early-stage breast cancer [5]. In 659 patients, ABUS combined with mammography demonstrated noninferior sensitivity to handheld ultrasound for detecting additional ipsilateral cancers and actually showed higher sensitivity for contralateral cancers. With similar specificity, the findings suggest ABUS is a feasible alternative to handheld ultrasound, which could be particularly valuable in settings with limited staffing or time constraints.

Our final theme covers papers that refine diagnostic pathways and offer new interpretation pearls. An international team of hand surgeons and musculoskeletal radiologists published interdisciplinary consensus statements on scaphoid fracture imaging in European Radiology [4]. Using a Delphi process, the experts agreed that radiographs should be the initial imaging technique. For suspected radiographically occult fractures, either MRI or CT is advocated. For assessing osseous consolidation, CT is the method of choice, while contrast-enhanced MRI is preferred for evaluating the vascularization of a scaphoid nonunion. These statements provide a clear and practical imaging pathway for a common clinical problem. In neuro-oncology, a study in the European Journal of Radiology addresses the common dilemma of sulcal enhancement on contrast-enhanced brain MRI [8]. Researchers found that comparing enhancement patterns on contrast-enhanced T2-FLAIR and T1-weighted images can reliably differentiate benign contrast leakage from true leptomeningeal metastasis. A mismatch between the two sequences was common in benign leakage, which was associated with older age and prior radiotherapy. Using T1-T2FLAIR matching as a criterion yielded a sensitivity of over 90% and specificity of nearly 95% for diagnosing metastasis, offering a simple and robust interpretation tool. Lastly, a powerful study in Nature provides fundamental new insights into the biology of diffuse midline gliomas, or DMGs [2]. Researchers used tumour network mapping to define a conserved brain network that is functionally connected to these devastating pediatric tumors. The degree of tumor connectivity within this network was an independent predictor of overall survival in two separate validation cohorts. The study also found that tumor growth mapped to these network trajectories and that incidental surgical resection of high-connectivity tissue conferred a significant survival advantage. This work supports the hypothesis that these gliomas exploit healthy brain circuits to promote their growth, opening new avenues for understanding and potentially treating this disease.

If you only have time for one paper this week, make it the study on unsedated non-contrast MRI for suspected pediatric appendicitis in the European Journal of Radiology [3]. It presents a practical, validated, and immediately applicable protocol that solves a common clinical challenge while completely avoiding ionizing radiation and the risks of sedation in children.

Here are the key takeaways from this week in Radiology. First, in the realm of artificial intelligence, general-purpose large language models are outperforming specialized clinical AI tools on knowledge-based tasks, while in image interpretation, human radiologists still hold a performance edge over current AI for critical diagnoses like cervical spine fractures. Second, in CT, you can substantially reduce radiation dose by about two-thirds in carotid CTA using low-dose protocols without sacrificing quantitative plaque analysis, and dual-energy CT with low-keV reconstructions significantly improves detection of residual viable HCC after TACE. Third, for children with suspected appendicitis, a rapid, non-contrast, unsedated MRI is a highly accurate and safe alternative to CT, capable of both ruling in or out appendicitis and identifying alternative diagnoses. Fourth, for preoperative staging in early-stage breast cancer, automated breast ultrasound is a noninferior alternative to handheld ultrasound, offering a potential solution to improve workflow and standardization. And finally, a simple MRI interpretation pearl: when evaluating sulcal enhancement in neuro-oncology patients, assessing for T1-T2FLAIR matching can reliably differentiate benign contrast leakage from true leptomeningeal metastasis.

That's your roundup for This Week in Radiology. The full transcript and references are available on the episode page in your AudioScholar library. This is an AI-curated summary — for clinical decisions, always consult primary sources and current guidelines. See you next week.

If this weekly briefing is useful, follow the show in your podcast app so new episodes arrive automatically. And for audio briefings on your own clinical questions and papers, visit audioscholar dot C C.

This is an automated summary generated by artificial intelligence, which can make mistakes. Always review the original source materials.