General Radiologists Research Articles

Enhancing chatbot performance for imaging recommendations: Leveraging GPT-4 and context-awareness for trustworthy clinical guidance

PurposeTo investigate if GPT-4 improves the accuracy, consistency, and trustworthiness of a context-aware chatbot to provide personalized imaging recommendations from American College of Radiology (ACR) appropriateness criteria documents using semantic similarity processing: In addition, we sought to enable auditability of the output by revealing the information source the decision relies on. Material and MethodsWe refined an existing chatbot that incorporated specialized knowledge of the ACR guidelines by upgrading GPT-3.5-Turbo to its successor GPT-4 by OpenAI, using the latest version of LlamaIndex, and improving the prompting strategy. This chatbot was compared to the previous version, generic GPT-3.5-Turbo and GPT-4, and general radiologists regarding the performance in applying the ACR appropriateness guidelines. ResultsThe refined context-aware chatbot performed superior to the previous version using GPT-3.5-Turbo, generic chatbots GPT-3.5-Turbo and GPT-4, and general radiologists in providing “usually or may be appropriate” recommendations according to the ACR guidelines (all p < 0.001). It also outperformed GPT-3.5-Turbo and general radiologists in respect to “usually appropriate” recommendations (both p < 0.001). Moreover, the consistency in correct answers was higher with 78 % consistent correct “usually appropriate” answers and 94 % for “usually or may be appropriate” recommendations. In all cases, the same source documents were chosen, ensuring transparency. ConclusionOur study demonstrates the significance of context awareness in ensuring the use of appropriate knowledge and proposes a strategy to enhance trust in chatbot-based outputs to provide transparency. The improvements in accuracy, consistency, and source transparency address trust issues and enhance the clinical decision support process.Abbreviations: ACR, American College of Radiology; accGPT, appropriateness criteria context aware GPT; accGPT-4, appropriateness criteria context aware GPT using GPT-4; GPT, generative pre-trained transformer; LLM, Large Language Model.

ESR Essentials: imaging of suspected child abuse-practice recommendations by the European Society of Paediatric Radiology.

The goal of this paper is to provide a useful desktop reference for the imaging of suspected child abuse with clear, age-specific pathways for appropriate evidence-based imaging and follow-up. We aim to provide a road map for the imaging evaluation and follow-up of this important and vulnerable cohort of patients presenting with signs and symptoms concerning forinflicted injury. As the imaging recommendations differ for children of different ages, we provide a flowchart of the appropriate imaging pathway for infants, toddlers, and older children, which allows ease of selection of which children should undergo skeletal survey, non-contrast computed tomography (CT) brain with 3-dimensional (D) reformats, and magnetic resonance imaging (MRI) of the brain and whole spine. For ease of review, we include a table of the common intracranial and spinal patterns of injury in abusive head trauma. We summarise search patterns, areas of review, and key findings to include in the report. To exclude skeletal injury, infants and children under 2 years of age should undergo a full skeletal survey in accordance with national guidelines, with a limited follow-up skeletal survey performed 11-14 days later. For children over 2 years of age, the need for skeletal imaging should be decided on a case-by-case basis. All infants should undergo a non-contrast-enhanced CT brain with 3-D reformats. If this is normal with no abnormal neurology, then no further neuroimaging is required. If this is abnormal, then they should proceed to MRI brain and whole spine within 2-5 days. Children older than 1 year of age who have abnormal neurology and/or findings on skeletal survey that are suggestive of inflicted injury should undergo non-contrast CT brain with 3-D reformats and, depending on the findings, may also require MRI of the brain and whole spine. We hope that this will be a helpful contribution to the radiology literature, particularly for the general radiologist with low volumes of paediatrics in their practice, supporting them with managing these important cases when they arise in daily practice. KEY POINTS: The choice of initial imaging (skeletal survey and/or brain CT) depends on the age of the child in whom abuse is suspected. A follow-up skeletal survey is mandatory 11-14 days after the initial survey. If an MRI of the brain is performed, then an MRI of the whole spine should be performed concurrently.

The Diagnostic Performance of Large Language Models and General Radiologists in Thoracic Radiology Cases: A Comparative Study.

To investigate and compare the diagnostic performance of 10 different large language models (LLMs) and 2 board-certified general radiologists in thoracic radiology cases published by The Society of Thoracic Radiology. We collected publicly available 124 "Case of the Month" from the Society of Thoracic Radiology website between March 2012 and December 2023. Medical history and imaging findings were input into LLMs for diagnosis and differential diagnosis, while radiologists independently visually provided their assessments. Cases were categorized anatomically (parenchyma, airways, mediastinum-pleura-chest wall, and vascular) and further classified as specific or nonspecific for radiologic diagnosis. Diagnostic accuracy and differential diagnosis scores (DDxScore) were analyzed using the χ2, Kruskal-Wallis, Wilcoxon, McNemar, and Mann-Whitney U tests. Among the 124 cases, Claude 3 Opus showed the highest diagnostic accuracy (70.29%), followed by ChatGPT 4/Google Gemini 1.5 Pro (59.75%), Meta Llama 3 70b (57.3%), ChatGPT 3.5 (53.2%), outperforming radiologists (52.4% and 41.1%) and other LLMs (P<0.05). Claude 3 Opus DDxScore was significantly better than other LLMs and radiologists, except ChatGPT 3.5 (P<0.05). All LLMs and radiologists showed greater accuracy in specific cases (P<0.05), with no DDxScore difference for Perplexity and Google Bard based on specificity (P>0.05). There were no significant differences between LLMs and radiologists in the diagnostic accuracy of anatomic subgroups (P>0.05), except for Meta Llama 3 70b in the vascular cases (P=0.040). Claude 3 Opus outperformed other LLMs and radiologists in text-based thoracic radiology cases. LLMs hold great promise for clinical decision systems under proper medical supervision.

Evaluating text and visual diagnostic capabilities of large language models on questions related to the Breast Imaging Reporting and Data System Atlas 5th edition.

This study aimed to evaluate the performance of large language models (LLMs) and multimodal LLMs in interpreting the Breast Imaging Reporting and Data System (BI-RADS) categories and providing clinical management recommendations for breast radiology in text-based and visual questions. This cross-sectional observational study involved two steps. In the first step, we compared ten LLMs (namely ChatGPT 4o, ChatGPT 4, ChatGPT 3.5, Google Gemini 1.5 Pro, Google Gemini 1.0, Microsoft Copilot, Perplexity, Claude 3.5 Sonnet, Claude 3 Opus, and Claude 3 Opus 200K), general radiologists, and a breast radiologist using 100 text-based multiple-choice questions (MCQs) related to the BI-RADS Atlas 5th edition. In the second step, we assessed the performance of five multimodal LLMs (ChatGPT 4o, ChatGPT 4V, Claude 3.5 Sonnet, Claude 3 Opus, and Google Gemini 1.5 Pro) in assigning BI-RADS categories and providing clinical management recommendations on 100 breast ultrasound images. The comparison of correct answers and accuracy by question types was analyzed using McNemar's and chi-squared tests. Management scores were analyzed using the Kruskal- Wallis and Wilcoxon tests. Claude 3.5 Sonnet achieved the highest accuracy in text-based MCQs (90%), followed by ChatGPT 4o (89%), outperforming all other LLMs and general radiologists (78% and 76%) (P < 0.05), except for the Claude 3 Opus models and the breast radiologist (82%) (P > 0.05). Lower-performing LLMs included Google Gemini 1.0 (61%) and ChatGPT 3.5 (60%). Performance across different categories of showed no significant variation among LLMs or radiologists (P > 0.05). For breast ultrasound images, Claude 3.5 Sonnet achieved 59% accuracy, significantly higher than other multimodal LLMs (P < 0.05). Management recommendations were evaluated using a 3-point Likert scale, with Claude 3.5 Sonnet scoring the highest (mean: 2.12 ± 0.97) (P < 0.05). Accuracy varied significantly across BI-RADS categories, except Claude 3 Opus (P < 0.05). Gemini 1.5 Pro failed to answer any BI-RADS 5 questions correctly. Similarly, ChatGPT 4V failed to answer any BI-RADS 1 questions correctly, making them the least accurate in these categories (P < 0.05). Although LLMs such as Claude 3.5 Sonnet and ChatGPT 4o show promise in text-based BI-RADS assessments, their limitations in visual diagnostics suggest they should be used cautiously and under radiologists' supervision to avoid misdiagnoses. This study demonstrates that while LLMs exhibit strong capabilities in text-based BI-RADS assessments, their visual diagnostic abilities are currently limited, necessitating further development and cautious application in clinical practice.

MR Imaging of Hemosiderin Deposition in the Ankle Joints of Patients with Haemophilia: The Contribution of a Multi-Echo Gradient-Echo Sequence-Correlation with Osteochondral Changes and the Number and Chronicity of Joint Bleeds.

We aim (a) to introduce an easy-to-perform multi-echo gradient-echo sequence (mGRE) for the detection of hemosiderin deposition in the ankle joints of boys with haemophilia (b) to explore the associations between the presence and severity of hemosiderin deposition and the other components of haemophilic arthropathy, the clinical score, and the number and chronicity of joint bleeds. An MRI of 41 ankle joints of 21 haemophilic boys was performed on a 3 T MRI system using an mGRE sequence in addition to the conventional protocol. Conventional MRI and mGRE were separately and independently assessed by three readers, namely, two musculoskeletal radiologists and a general radiologist for joint hemosiderin. We set as a reference the consensus reading of the two musculoskeletal radiologists, who also evaluated the presence of synovial thickening, effusion, and osteochondral changes. Excellent inter-reader agreement was obtained using the mGRE sequence compared to the conventional protocol (ICC: 0.95-0.97 versus 0.48-0.89), with superior sensitivity (90-95% versus 50-85%), specificity (95.2-100% versus 76.2-95.2%), and positive (95-100% versus 71-94.4%) and negative predictive value (91.3-95.5% versus 87-63%). Hemosiderin deposition was associated with osteochondral changes, synovial thickening, clinical score, and the total number of ankle bleeds, while it was inversely related with the time elapsed between the last joint bleed and MRI. (p < 0.05). The application of an mGRE sequence significantly improved hemosiderin detection, even when performed by the less experienced reader. Joint hemosiderin deposition was associated with the other components of haemophilic arthropathy and was mostly apparent in recent joint bleeds.

Comparative analysis of GPT-4-based ChatGPT's diagnostic performance with radiologists using real-world radiology reports of brain tumors.

Large language models like GPT-4 have demonstrated potential for diagnosis in radiology. Previous studies investigating this potential primarily utilized quizzes from academic journals. This study aimed to assess the diagnostic capabilities of GPT-4-based Chat Generative Pre-trained Transformer (ChatGPT) using actual clinical radiology reports of brain tumors and compare its performance with that of neuroradiologists and general radiologists. We collected brain MRI reports written in Japanese from preoperative brain tumor patients at two institutions from January 2017 to December 2021. The MRI reports were translated into English by radiologists. GPT-4 and five radiologists were presented with the same textual findings from the reports and asked to suggest differential and final diagnoses. The pathological diagnosis of the excised tumor served as the ground truth. McNemar's test and Fisher's exact test were used for statistical analysis. In a study analyzing 150 radiological reports, GPT-4 achieved a final diagnostic accuracy of 73%, while radiologists' accuracy ranged from 65 to 79%. GPT-4's final diagnostic accuracy using reports from neuroradiologists was higher at 80%, compared to 60% using those from general radiologists. In the realm of differential diagnoses, GPT-4's accuracy was 94%, while radiologists' fell between 73 and 89%. Notably, for these differential diagnoses, GPT-4's accuracy remained consistent whether reports were from neuroradiologists or general radiologists. GPT-4 exhibited good diagnostic capability, comparable to neuroradiologists in differentiating brain tumors from MRI reports. GPT-4 can be a second opinion for neuroradiologists on final diagnoses and a guidance tool for general radiologists and residents. This study evaluated GPT-4-based ChatGPT's diagnostic capabilities using real-world clinical MRI reports from brain tumor cases, revealing that its accuracy in interpreting brain tumors from MRI findings is competitive with radiologists. We investigated the diagnostic accuracy of GPT-4 using real-world clinical MRI reports of brain tumors. GPT-4 achieved final and differential diagnostic accuracy that is comparable with neuroradiologists. GPT-4 has the potential to improve the diagnostic process in clinical radiology.

Remote Teleproctoring with the TEGUS System for Mechanical Thrombectomy in aNon-Comprehensive Stroke Center: Initial Preliminary Data On Clinical Experience.

Mechanical thrombectomy (MT) is typically performed by experienced neurointerventional radiologists. However, logistical and geographic limitations often hinder access to rapid MT. This study reports the first clinical experience using TEGUS teleproctoring to support MT conducted by general interventional radiologists (IR) at non-comprehensive stroke centers, compared to on-site proctoring outcomes. The Arnau de Vilanova Hospital in Spain used to transfer stroke patients requiring MT to acomprehensive Stroke Center 160 km away. To overcome COVID-19 mobility restrictions, the Tegus Teleproctoring System was installed. Before teleproctoring, the general interventional radiologist underwent six months of neurointerventional training at aprimary stroke center. From April 2021 to May 2023, general IR conducted MT either with on-site proctor supervision or teleproctoring support. We aim to compare clinical outcome of patients receiving MT according to proctoring method. During the study, 49MTs were performed: 15 with TEGUS teleproctoring and 34with on-site proctoring. Both groups had similar baseline characteristics, except for NIHSS scores (Tegus 9 [IQR 6-20] vs 18 [IQR 12-22], p = 0.034). No significant differences were found in door-to-revascularization time (82 ± 28.2 vs 84 ± 26.4) min, p = (0.895). The final mTICI distribution and 90-day mRS scores were comparable after adjusting by stroke severity. There were no reports of symptomatic intracranial hemorrhage in either group. This study shows the feasibility of Tegus remote teleproctoring during emergent cases of MT in aremote hospital. It could improve the learning curve of interventional radiologists with limited experience in MT, and lower the territorial inequity associated to MT.

An approach leveraging radiomics and model checking for the automatic early diagnosis of adhesive capsulitis

Adhesive Capsulitis of the shoulder is a painful pathology limiting shoulder movements, commonly known as “Frozen Shoulder”. Since this pathology limits movement, it is important to make an early diagnosis. Diagnosing capsulitis relies on clinical assessment, although diagnostic imaging, such as Magnetic Resonance Imaging, can provide predictive or supportive information for specific characteristic signs. However, its diagnosis is not so simple nor so immediate, indeed it remains a difficult topic for many general radiologists and expert musculoskeletal radiologists. This study aims to investigate whether it is possible to use disease signs within a medical image to automatically diagnose Adhesive Capsulitis. To this purpose, we propose an automatic Model Checking-based approach to quickly diagnose the Adhesive Capsulitis taking as input the radiomic feature values from the medical images. Furthermore, we compare the performance achieved by our method with diagnostic results obtained by professional radiologists with different levels of experience. To the best of our knowledge, this is the first method for the automatic diagnosis of Adhesive Capsulitis of the Shoulder.

ESR Essentials: response assessment criteria in oncologic imaging-practice recommendations by the European Society of Oncologic Imaging.

Assessing the response to oncological treatments is paramount for determining the prognosis and defining the best treatment for each patient. Several biomarkers, including imaging, can be used, but standardization is fundamental for consistency and reliability. Tumor response evaluation criteria have been defined by international groups for application in pharmaceutical clinical trials evaluating new drugs or therapeutic strategies. RECIST 1.1 criteria are exclusively based on unidimensional lesion measurements; changes in tumor size are used as surrogate imaging biomarkers to correlate with patient outcomes. However, increased tumor size does not always reflect tumor progression. The introduction of immunotherapy has led to the development of new criteria (iRECIST, Level of Evidence (LoE) Ib) that consider the possibility that an increase in disease burden is secondary to the immune response instead of progression, with the new concept of Unconfirmed Progressive Disease (a first progression event which must be confirmed on follow-up). Specific criteria were devised for HCC (mRECIST, LoE IV), which measure only enhancing HCC portions to account for changes after local therapy. For GIST treated with imatinib, criteria were developed to account for the possible increase in size reflecting a response rather than a progression by assessing both tumor size and density on CT (Choi, LoE II). This article provides concise and relevant practice recommendations aimed at general radiologists to help choose and apply the most appropriate criteria for assessing response to treatment in different oncologic scenarios. Though these criteria were developed for clinical trials, they may be applied in clinical practice as a guide for day-to-day interpretation. KEY POINTS: Response evaluation criteria, designed for use in clinical trials, might serve as a surrogate biomarker for overall survival. RECIST 1.1 defines measurable and non-measurable disease among which target lesions and non-target lesions are selected at baseline as reference for follow-ups. Some therapies and/or cancers require the use of different criteria, such as iRECIST, mRECIST, and Choi criteria.

Analysis of Requirements within Neuroradiology Job Advertisements in a Specific Timeframe.

The neuroradiology job market is constantly changing along with the skill sets needed by fellowship graduates to participate successfully in the job market. This study aimed to establish a baseline of employer requirements in the neuroradiology job market within the study timeframe. The American Society of Neuroradiology and American College of Radiology job boards were queried for neuroradiology positions between August 12, 2022, and December 31, 2022. The positions and requirements were categorized into academic versus private practice, general diagnostic radiology, full-subspecialized neuroradiology, hybrid remote/onsite, outpatient, inpatient/emergency, general interventional radiology procedures, and neuroradiology procedures. Exclusion criteria included neurointerventional only, remote-only, pediatric-only, no preference for neuroradiology, and duplicate posts within and between the job boards. Of 1777 total job posts, 179 were neuroradiology-specific and the remainder were general. Of the 179 neuroradiology-specific jobs, 55 neuroradiology jobs were academic and 124 jobs were private practice. A higher proportion of private practice jobs required general diagnostic interpretations (83% versus 26%), a higher proportion of academic jobs required neuroradiology procedures (56% versus 31%), and a higher proportion of private practice jobs required general interventional radiology procedures (22% versus 0%). Thirty-nine percent of all neuroradiology-specific onsite jobs required neuroradiology procedures, and 15% required general interventional radiology procedures. Because there was a sizable difference between general radiology and procedure requirements between academic and private practice positions, tailoring fellowship training for career aspirations of neuroradiology fellows should be considered to adapt to the skills needed for the evolving job market. In the queried timeframe, 61% of neuroradiology-specific onsite jobs did not have a specific procedure skill requirement for job applicants. This article serves as a single snapshot of the job market and its requirements for neuroradiologists, to aid in planning training to meet the needs of employers.

Does delayed phase imaging in CT angiography provide additional information in patients with suspected active bleeding?

We hypothesize that delayed phase imaging does not provide additional diagnostic information in patients who undergo multi-phasic CTA for suspected active bleeding. Data on patients who underwent multiphasic CTA (pre-contrast, arterial, porto-venous, and delayed phases) for suspected acute bleed were retrospectively collected between January 2019 and November 2021. CTA images were reviewed by a general radiologist, an interventional radiologist, and a body imaging radiologist independently. Each reader evaluated if delayed phase images provided additional information that would change the final impression of the CTA report. Additional information regarding bleeding location, time needed for delayed image acquisition, and radiation exposure were also obtained. A total of 104 patients with CTAs were analyzed with an average age of 58 years ± 22. Studies rated with absent additional findings on delayed images were 102 (98.1%) by the interventional radiologist, 101 (97.1%) by the body imaging radiologist, and 100 (96.1%) by the general radiologist with percent agreement of 96.15% (kappa 0.54, p < 0.001). All the findings were characterized as unlikely to be clinically significant. Mean time added to complete a delayed phase images was 3.61 ± 3.4min. The average CT dose length product (DLP) for the total exam was 3621.78 ± 2129.57 mGy.cm with delayed acquisition adding a mean DLP of 847.75 ± 508.8 mGy.cm. Delayed phase imaging does not provide significant additional diagnostic information in evaluating patients with suspected active bleeding but is associated with increased examination time and radiation exposure.

Standardized reporting of spine and sacroiliac joints in axial spondyloarthritis MRI: from the ESSR-Arthritis Subcommittee.

Apply a modified Delphi-based approach and produce a practical, radiology-specific set of definitions for interpretation and standardization of the multiple MRI findings in axial spondyloarthritis (ax-SpA), specifically to aid the general radiologist with a musculoskeletal interest, working with gold standard basic MRI protocols. We report the results of a modified Delphi-based consensus of 35 experts from 13 countries in the Arthritis Subcommittee of the European Society of Musculoskeletal Radiology (ESSR). Seventeen definitions were created (i.e., nine for the spine and eight for the sacroiliac joint) and two Delphi rounds were conducted on an electronic database, collated and revised by the project leader with agreement. Group leads were appointed for each definition following the first round. Final definitions included only those that reached a consensus > 80%; if > 50% agreed on exclusion consensus, definitions were excluded. Final results have been shared during the Arthritis meeting at the Annual ESSR Congress. Fourteen definitions, eight for the spine and six for the sacroiliac joint were agreed for standardized reporting. Andersson's, anterior corner sclerotic and costovertebral joint inflammatory lesions of the spine, with active and non-active erosions, and fat metaplasia of the sacroiliac joint reaching the highest consensus (≥ 95%). More than 50% of the experts agreed to exclude joint space inflammation in the sacroiliac joint and tissue backfill. Syndesmophytes reached 76% agreement. Agreed definitions by expert radiologists using a modified Delphi process, should allow standardized actionable radiology reports and clarity in reporting terminology of ax-SpA. The proposed definitions will support reporting from musculoskeletal and general radiologists working with gold-standard basic MRI, improve confidence in lesion assessment, and standardize terminology to provide actionable reports on MRI in patients with ax-SpA. Experts applied a modified Delphi method to optimize the definitions of MRI findings of ax-SpA. After two Delphi rounds and one in-person meeting, fourteen definitions reached the agreement threshold. These consensus-based definitions will aid in actionable reporting specifically for the general radiologist with a musculoskeletal interest.

Associations between 24/7 workforce coverage, organizational challenges, and well-being of emergency radiologists: a national survey.

The last decade has seen a surge in the demand for imaging exams in emergency radiology (ER), necessitating an evolution in organizational systems for departments offering round-the-clock care, while safeguarding patient care quality and physician well-being to prevent burnout. To develop a nationwide overview of ER organizations in France and identify structures that promote job satisfaction. Two surveys were sent to 709 radiological centers across France from March to June 2022, inquiring about organizational aspects and quality of life (QOL), incorporating four validated QOL questionnaires. The organization of each center was mapped, and correlations between respondent characteristics and mental health were analyzed using Pearson's and Wilcoxon tests. A total of 284 centers answered the organizational survey, with a response rate of about 41.6%. Among them, there were 32 university hospitals, 208 general hospitals, 2 teaching army hospitals, and 42 private facilities. Of these, night-time operations showed 14% on-site coverage, 12% on-call from home, 69% utilized external teleradiology, and 4% used in-house teleradiology. These trends persisted over weekends and holidays. Regarding the quality of working life, academic, general, and private radiologists are more satisfied with their practice compared to trainees. Depersonalization, part of the three dimensions of burnout, was high in every class, at 60% (n = 210/350). Outside of university hospitals, most radiology centers in France no longer have on-site radiologists during off hours. Residents are prone to lower job satisfaction and quality of life than more experienced radiologists. The survey illustrates how French ER is structured, pointing out the escalating significance of teleradiology and noting that radiologists generally experience high job satisfaction while also confronting typical organizational challenges. The need for continuous radiology coverage comes with unique logistical challenges, especially in ER. Night shifts show a significant reliance on teleradiology services, especially by external companies. Pay, shift patterns, and seniority affect the well-being of emergency radiologists, particularly the residents.

Merlin: A Vision Language Foundation Model for 3D Computed Tomography.

Over 85 million computed tomography (CT) scans are performed annually in the US, of which approximately one quarter focus on the abdomen. Given the current shortage of both general and specialized radiologists, there is a large impetus to use artificial intelligence to alleviate the burden of interpreting these complex imaging studies while simultaneously using the images to extract novel physiological insights. Prior state-of-the-art approaches for automated medical image interpretation leverage vision language models (VLMs) that utilize both the image and the corresponding textual radiology reports. However, current medical VLMs are generally limited to 2D images and short reports. To overcome these shortcomings for abdominal CT interpretation, we introduce Merlin - a 3D VLM that leverages both structured electronic health records (EHR) and unstructured radiology reports for pretraining without requiring additional manual annotations. We train Merlin using a high-quality clinical dataset of paired CT scans (6+ million images from 15,331 CTs), EHR diagnosis codes (1.8+ million codes), and radiology reports (6+ million tokens) for training. We comprehensively evaluate Merlin on 6 task types and 752 individual tasks. The non-adapted (off-the-shelf) tasks include zero-shot findings classification (31 findings), phenotype classification (692 phenotypes), and zero-shot cross-modal retrieval (image to findings and image to impressions), while model adapted tasks include 5-year chronic disease prediction (6 diseases), radiology report generation, and 3D semantic segmentation (20 organs). We perform internal validation on a test set of 5,137 CTs, and external validation on 7,000 clinical CTs and on two public CT datasets (VerSe, TotalSegmentator). Beyond these clinically-relevant evaluations, we assess the efficacy of various network architectures and training strategies to depict that Merlin has favorable performance to existing task-specific baselines. We derive data scaling laws to empirically assess training data needs for requisite downstream task performance. Furthermore, unlike conventional VLMs that require hundreds of GPUs for training, we perform all training on a single GPU. This computationally efficient design can help democratize foundation model training, especially for health systems with compute constraints. We plan to release our trained models, code, and dataset, pending manual removal of all protected health information.

Optimizing Advanced Imaging of the Pediatric Patient in the Emergency Department: Policy Statement

Advanced imaging, including ultrasonography, computed tomography, and magnetic resonance imaging (MRI), is an integral component to the evaluation and management of ill and injured children in the emergency department. As with any test or intervention, the benefits and potential impacts on management must be weighed against the risks to ensure that high-value care is being delivered. There are important considerations specific to the pediatric patient related to the ordering and interpretation of advanced imaging. This policy statement provides guidelines for institutions and those who care for children to optimize the use of advanced imaging in the emergency department setting and was coauthored by experts in pediatric and general emergency medicine, pediatric radiology, and pediatric surgery. The intent is to guide decision-making where children may access care.

Optimizing Advanced Imaging of the Pediatric Patient in the Emergency Department: Policy Statement.

Advanced imaging, including ultrasonography, computed tomography, and magnetic resonance imaging, is an integral component to the evaluation and management of ill and injured children in the emergency department. As with any test or intervention, the benefits and potential impacts on management must be weighed against the risks to ensure that high-value care is being delivered. There are important considerations specific to the pediatric patient related to the ordering and interpretation of advanced imaging. This policy statement provides guidelines for institutions and those who care for children to optimize the use of advanced imaging in the emergency department setting and was coauthored by experts in pediatric and general emergency medicine, pediatric radiology, and pediatric surgery. The intent is to guide decision-making where children may access care.

Stopping anticoagulation for isolated or incidental pulmonary embolism: the STOPAPE RCT protocol.

Is withholding anticoagulation for patients with isolated or incidental subsegmental pulmonary embolism clinically and cost-effective compared with full anticoagulation for 3 months? There has been an increase in the diagnosis of subsegmental pulmonary embolism since the advent of computed tomography pulmonary angiogram to investigate patients with suspected pulmonary embolism. Subsegmental pulmonary embolism is not often detectable with older nuclear medicine-based diagnostic imaging for ventilation/perfusion mismatch. The case fatality of pulmonary embolism has reduced as subsegmental pulmonary embolism diagnoses from computed tomography pulmonary angiogram have increased. There is growing equipoise about the optimal treatment for patients with subsegmental pulmonary embolism, given that full anticoagulation has significant risks of bleeding and subsegmental pulmonary embolism was not often diagnosed previously with ventilation/perfusion scanning and therefore most likely left predominantly untreated prior to the introduction of computed tomography pulmonary angiogram scanning. Determine whether withholding anticoagulation for isolated or incidental subsegmental pulmonary embolism (i.e. subsegmental pulmonary embolism with no coexisting deep-vein thrombosis) reduces the harms of recurrent thromboembolism and major bleeding compared with 3 months of full anticoagulation at 3, 6 and 12 months. Determine the rate of complications of anticoagulation therapy (predominantly bleeding) in patients with isolated subsegmental pulmonary embolism. Determine whether not treating isolated subsegmental pulmonary embolism is acceptable to clinicians and patients. Determine the reclassification rate of subsegmental pulmonary embolism diagnoses made by general reporting radiologists when reviewed by specialist respiratory radiologists and develop a set of rules to improve general radiologists' diagnoses of subsegmental pulmonary embolism. Assess cost-effectiveness of not treating patients with isolated subsegmental pulmonary embolism with anticoagulation, taking a health service perspective. Prospective individually randomised open controlled trial with blinded end-point committee assessment for outcomes, powered for non-inferiority for recurrent venous thromboembolism and for superiority for bleeding events. An internal pilot phase is included for feasibility and acceptability of no anticoagulation. We planned to recruit 1466 patients from at least 50 acute hospital sites. Allowing for a dropout rate of 15%, this would have given us 90% power to detect a reduction in major and clinically relevant non-major bleeding from 7.3% in the anticoagulation arm to 3% in the intervention arm. We were powered to determine that a strategy of no anticoagulation was non-inferior to anticoagulation with an upper margin of a 2.3% increase in recurrent venous thromboembolism from an expected rate of 2% in those who receive full anticoagulation. We also planned to undertake a study comparing acute reporting radiologists' diagnoses of subsegmental pulmonary embolism from all computed tomography pulmonary angiograms with specialist respiratory radiologists. This would have allowed us to determine safety in the pilot study (i.e. patients with pulmonary embolism that was in fact larger than subsegmental would have been identified) and develop guidance for subsegmental pulmonary embolism diagnosis for general radiologists. Patients with lived experience of thrombosis contributed to all aspects of the trial design and were part of the Trial Management Group. The STOPAPE trial was stopped prematurely due to a low recruitment rate in the wake of the COVID pandemic and prioritisation of recovery of the National Institute for Health and Care Research research portfolio. There are no outcome data available for this trial. Separate NIHR Library publications will detail the linked qualitative study examining the views of patients and clinicians around withholding anticoagulation for isolated subsegmental pulmonary embolism as well as presenting all collected data of recruited patients. This article presents independent research funded by the National Institute for Health and Care Research (NIHR) Health Technology Assessment programme as award number NIHR128073. A plain language summary of this research article is available on the NIHR Journals Library Website https://doi.org/10.3310/HRCW7937.

Imaging Features of Immunodeficiency-Associated Primary CNS Lymphoma: A Review.

In the immunocompromised setting, there are distinct radiologic findings of primary central nervous system lymphoma (PCNSL), including necrotic ring-enhancing lesions, increased propensity for intralesional haemorrhage, and multiplicity. In this clinical context, advanced imaging with MR perfusion, spectroscopy, and diffusion-weighted imaging can be used to increase accuracy in the diagnosis of lymphoma over mimics such as high-grade glioma, metastases, or infection. This review summarizes the histology and pathophysiology of PCNSL in immunodeficient hosts, which provide a basis for its imaging appearances, prognosis, and treatment. This discussion is important for the general radiologist as the incidence of immunodeficiency-related PCNSL may be increasing.

Selective ophthalmic arterial injection using a balloon catheter for retinoblastoma: a seven-year clinical evaluation.

To evaluate the effectiveness and safety of selective ophthalmic arterial injection (SOAI) for retinoblastoma utilizing a microballoon catheter system with an M chamber. Retrospective analysis. This study was sanctioned by theNational Cancer Center Hospital' Independent Ethics Committee. The surgeon was a general interventional radiologist. After confirming that the distal internal carotid artery was not delineated by balloon occlusion and the ophthalmic artery was visualized using digital subtraction angiography, melphalan was manually administered. Notably, in cases presenting bilateral retinoblastoma, both eyes received treatment in a singular, low-dose procedure. Between July 2015 and December 2021, 125 patients with retinoblastoma (68 boys and 57 girls) underwent SOAI at our facility. The average age at initial treatment was 19.3 months. The study covered 250 procedures, with patients undergoing an average of 3.7 procedures. The success rate of the procedure was 99.2%, with a mean procedure duration of 18.3min. Two distinct technical failures were recorded: one attributed to an internal carotid artery having a wide lumen and the other due to the ophthalmic artery remaining undetected on angiography post-balloon occlusion of the internal carotid artery. Adverse events were minimal but included bronchospasm post-procedure and severe orbital inflammation in 0.8% and 0.4% of cases, respectively. SOAI using the microballoon catheter with the M chamber is a feasible and safe procedure for the treatment of retinoblastoma. The success rate was 99.2%. This system can be recommended as intra-arterial chemotherapy for retinoblastoma.

Mammographic density assessment: comparison of radiologists, automated volumetric measurement, and artificial intelligence-based computer-assisted diagnosis.

Artificial intelligence-based computer-assisted diagnosis (AI-CAD) is increasingly used for mammographic exams, and its role in mammographic density assessment should be evaluated. To assess the inter-modality agreement between radiologists, automated volumetric density measurement program (Volpara), and AI-CAD system in breast density categorization using the Breast Imaging-Reporting and Data System (BI-RADS) density categories. A retrospective review was conducted on 1015 screening digital mammograms that were performed in Asian female patients (mean age = 56 years ± 10 years) in our health examination center between December 2022 and January 2023. Four radiologists with two different levels of experience (expert and general radiologists) performed density assessments. Agreement between the radiologists, Volpara, and AI-CAD (Lunit INSIGHT MMG) was evaluated using weighted kappa statistics and matched rates. Inter-reader agreement between expert and general radiologists was substantial (k = 0.65) with a matched rate of 72.8%. The agreement was substantial between expert or general radiologists and Volpara (k = 0.64-0.67) with a matched rate of 72.0% but moderate between expert or general radiologists and AI-CAD (k = 0.45-0.58) with matched rates of 56.7%-67.0%. The agreement between Volpara and AI-CAD was moderate (k = 0.53) with a matched rate of 60.8%. The agreement in breast density categorization between radiologists and automated volumetric density measurement program (Volpara) was higher than the agreement between radiologists and AI-CAD (Lunit INSIGHT MMG).