Radiology最新文献

Background Application of multimodal large language models (LLMs) with both textual and visual capabilities has been steadily increasing, but their ability to interpret radiologic images is still doubted. Purpose To evaluate the accuracy of LLMs and compare it with that of human readers with varying levels of experience and to assess the factors affecting LLM accuracy in answering New England Journal of Medicine Image Challenge cases. Materials and Methods Radiologic images of cases from October 13, 2005, to April 18, 2024, were retrospectively reviewed. Using text and image inputs, LLMs (Open AI's GPT-4 Turbo with Vision [GPT-4V] and GPT-4 Omni [GPT-4o], Google's DeepMind Gemini 1.5 Pro, and Anthropic's Claude 3) provided answers. Human readers (seven junior faculty radiologists, two clinicians, one in-training radiologist, and one medical student), blinded to the published answers, also answered. LLM accuracy with and without image inputs and short (cases from 2005 to 2015) versus long text inputs (from 2016 to 2024) was evaluated in subgroup analysis to determine the effect of these factors. Factor analysis was assessed using multivariable logistic regression. Accuracy was compared with generalized estimating equations, with multiple comparisons adjusted by using Bonferroni correction. Results A total of 272 cases were included. GPT-4o achieved the highest overall accuracy among LLMs (59.6%; 162 of 272), outperforming a medical student (47.1%; 128 of 272; P < .001) but not junior faculty (80.9%; 220 of 272; P < .001) or the in-training radiologist (70.2%; 191 of 272; P = .003). GPT-4o exhibited similar accuracy regardless of image inputs (without images vs with images, 54.0% [147 of 272] vs 59.6% [162 of 272], respectively; P = .59). Human reader accuracy was unaffected by text length, whereas LLMs demonstrated higher accuracy with long text inputs (all P < .001). Text input length affected LLM accuracy (odds ratio range, 3.2 [95% CI: 1.9, 5.5] to 6.6 [95% CI: 3.7, 12.0]). Conclusion LLMs demonstrated substantial accuracy with text and image inputs, outperforming a medical student. However, their accuracy decreased with shorter text lengths, regardless of image input. © RSNA, 2024 Supplemental material is available for this article.

History: A 10-month-old female infant, who was second-born, was referred for progressive macrocephaly, axial hypotonia, developmental delay, and limb stiffness. Birth had occurred at 41 weeks, after an uneventful pregnancy and delivery, to nonconsanguineous parents. Noticeably, the child could not hold her head up at 4 months or sit at 10 months of age. Her vocalizations were modulated to express contentment or anger, without developmentally appropriate babbling. Neurologic examination revealed poor visual contact, progressive macrocephaly, substantial axial hypotonia, and limb stiffness with brisk osteotendinous reflexes, suggestive of spastic diplegia. No swallowing difficulties or seizures were reported, and long-term electroencephalographic monitoring revealed no abnormalities. The patient underwent 3-T MRI (Siemens Healthineers) of the brain, including morphologic sequences and spectroscopy (Figs 1-4), under general anesthesia with and without gadolinium-based contrast media administration (Dotarem; Guerbet).

Background The detection of in-stent restenosis (ISR) with coronary CT angiography (CCTA) is challenging, but CT perfusion (CTP) has demonstrated improved diagnostic accuracy over CCTA in patients with stents. However, there are limited data on the performance of dynamic CTP, which allows noninvasive adjudication of regional myocardial blood flow. Purpose To compare the diagnostic performance of regadenoson-stress dynamic CTP with that of CCTA, using fractional flow reserve (FFR) and the index of microvascular resistance (IMR) as reference standards for epicardial coronary circulation and coronary microcirculation, respectively. Materials and Methods Between January 2021 and June 2022, this prospective study enrolled patients with stents with indication for invasive coronary angiography due to suspicion of ISR or coronary artery disease progression. Participants underwent dynamic stress myocardial CTP and rest CTP plus CCTA. A wide coverage (z-axis coverage, 16 cm) and fast (gantry rotation time, 0.28 second) scanner was used. During invasive coronary angiography, FFR and IMR were obtained. The diagnostic rate (number of interpretable territories divided by number of evaluated territories) and accuracy of CCTA and CTP were evaluated in a territory-based analysis and compared with FFR and IMR (primary end points of the study). Results The study included 156 consecutive patients (136 men [87%]; mean age, 63.1 years ± 8.2 [SD]) with 504 stents. The diagnostic rate was higher for CTP than for CCTA (98.7% [789 of 799 territories] vs 95.6% [764 of 799 territories], P < .001). With use of FFR as the reference standard, sensitivity, specificity, and diagnostic accuracy were higher for CTP than for CCTA (89.0%, 82.8%, and 84.7%, respectively, vs 60.0%, 61.9%, and 61.5%; P < .001). With use of IMR as the reference standard, sensitivity, specificity, and diagnostic accuracy were higher with CTP than with CCTA (76.5%, 85.9%, and 82.9%, respectively, vs 48.2%, 63.5%, and 59.3%; P < .01). The mean effective dose of stress CTP plus CCTA was 10.4 mSv ± 2.7. Conclusion In patients with coronary stents, dynamic CTP improves the diagnostic performance of CCTA in the detection of territory-based ischemia. © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Williams in this issue.

Background An accurate method of assessing the response of muscle-invasive bladder cancer (MIBC) to neoadjuvant treatment is needed for selecting candidates for bladder-sparing strategies. Purpose To evaluate the diagnostic accuracy and reproducibility of neoadjuvant chemotherapy Vesical Imaging Reporting and Data System (nacVI-RADS) scores and posttreatment Vesical Imaging Reporting and Data System (VI-RADS) scores when assessing MIBC response to neoadjuvant immunotherapy with multiparametric MRI (mpMRI). Materials and Methods A retrospective analysis of MRI scans was conducted in patients enrolled in the PURE-01 study (NCT02736266) from February 2017 to December 2019 who underwent pre- and postimmunotherapy mpMRI before radical cystectomy. Five readers independently reviewed the scans using VI-RADS and nacVI-RADS criteria. Diagnostic accuracy was evaluated for each reader, and the final histopathologic diagnosis served as the reference standard. Interreader agreement was assessed with the percentage of agreement, Conger κ, and Gwet agreement coefficient AC1. Results A total of 110 patients (median age, 67 years [IQR: 61-74]; 96 male) with 220 MRI scans were included; 80 (73%) patients had pure urothelial carcinoma. A total of 46 of 110 (42%) patients achieved a complete pathologic response. The sensitivity, specificity, and negative predictive value of nacVI-RADS 3 or higher for detecting residual disease (higher than stage ypT0) at radical cystectomy were 67%-84%, 63%-96%, and 63%-75%, respectively; for residual muscle-invasive disease (higher than stage ypT1), these values were 91%-98%, 55%-94%, and 93%-98%, respectively. The accuracy of nacVI-RADS was 72%-81% for stage ypT0 or higher disease and 71%-95% for stage ypT1 or higher disease. The accuracy of VI-RADS 3 or higher was 80%-95% for stage ypT1 or higher disease. The percentage of agreement for nacVI-RADS scores was 82% (κ = 0.62-0.65; AC1 = 0.65). Conclusion The nacVI-RADS scores showed good accuracy and reproducibility when assessing MIBC response to neoadjuvant immunotherapy. © RSNA, 2024 Supplemental material is available for this article.

Background High-quality translations of radiology reports are essential for optimal patient care. Because of limited availability of human translators with medical expertise, large language models (LLMs) are a promising solution, but their ability to translate radiology reports remains largely unexplored. Purpose To evaluate the accuracy and quality of various LLMs in translating radiology reports across high-resource languages (English, Italian, French, German, and Chinese) and low-resource languages (Swedish, Turkish, Russian, Greek, and Thai). Materials and Methods A dataset of 100 synthetic free-text radiology reports from CT and MRI scans was translated by 18 radiologists between January 14 and May 2, 2024, into nine target languages. Ten LLMs, including GPT-4 (OpenAI), Llama 3 (Meta), and Mixtral models (Mistral AI), were used for automated translation. Translation accuracy and quality were assessed with use of BiLingual Evaluation Understudy (BLEU) score, translation error rate (TER), and CHaRacter-level F-score (chrF++) metrics. Statistical significance was evaluated with use of paired t tests with Holm-Bonferroni corrections. Radiologists also conducted a qualitative evaluation of translations with use of a standardized questionnaire. Results GPT-4 demonstrated the best overall translation quality, particularly from English to German (BLEU score: 35.0 ± 16.3 [SD]; TER: 61.7 ± 21.2; chrF++: 70.6 ± 9.4), to Greek (BLEU: 32.6 ± 10.1; TER: 52.4 ± 10.6; chrF++: 62.8 ± 6.4), to Thai (BLEU: 53.2 ± 7.3; TER: 74.3 ± 5.2; chrF++: 48.4 ± 6.6), and to Turkish (BLEU: 35.5 ± 6.6; TER: 52.7 ± 7.4; chrF++: 70.7 ± 3.7). GPT-3.5 showed highest accuracy in translations from English to French, and Qwen1.5 excelled in English-to-Chinese translations, whereas Mixtral 8x22B performed best in Italian-to-English translations. The qualitative evaluation revealed that LLMs excelled in clarity, readability, and consistency with the original meaning but showed moderate medical terminology accuracy. Conclusion LLMs showed high accuracy and quality for translating radiology reports, although results varied by model and language pair. © RSNA, 2024 Supplemental material is available for this article.