Search

Purpose: Large language models (LLMs) like Generative Pre-trained Transformer 4 (GPT-4) can assist in detecting errors in radiology reports, but privacy concerns limit their clinical applicability. This study compares closed-source and privacy-compliant open-source LLMs for detecting common errors in radiology reports., Materials and Methods: A total of 120 radiology reports were compiled (30 each from X-ray, ultrasound, CT, and MRI). Subsequently, 397 errors from five categories (typographical, numerical, findings-impression discrepancies, omission/insertion, interpretation) were inserted into 100 of these reports; 20 reports were left unchanged. Two open-source models (Llama 3-70b, Mixtral 8x22b) and two commercial closed-source (GPT-4, GPT-4o) were tasked with error detection using identical prompts. The Kruskall-Wallis test and paired t-test were used for statistical analysis., Results: Open-source LLMs required less processing time per radiology report than closed-source LLMs (6 ± 2 s vs. 13 ± 4 s; p < 0.001). Closed-source LLMs achieved higher error detection rates than open-source LLMs (GPT-4o: 88% [348/397; 95% CI: 86, 92], GPT-4: 83% [328/397; 95% CI: 80, 87], Llama 3-70b: 79% [311/397; 95% CI: 76, 83], Mixtral 8x22b: 73% [288/397; 95% CI: 68, 77]; p < 0.001). Numerical errors (88% [67/76; 95% CI: 82, 93]) were detected significantly more often than typographical errors (75% [65/86; 95% CI: 68, 82]; p = 0.02), discrepancies between findings and impression (73% [73/101; 95% CI: 67, 80]; p < 0.01), and interpretation errors (70% [50/71; 95% CI: 62, 78]; p = 0.001)., Conclusion: Open-source LLMs demonstrated effective error detection, albeit with comparatively lower accuracy than commercial closed-source models, and have potential for clinical applications when deployed via privacy-compliant local hosting solutions., Key Points: Question Can privacy-compliant open-source large language models (LLMs) match the error-detection performance of commercial non-privacy-compliant closed-source models in radiology reports? Findings Closed-source LLMs achieved slightly higher accuracy in detecting radiology report errors than open-source models, with Llama 3-70b yielding the best results among the open-source models. Clinical relevance Open-source LLMs offer a privacy-compliant alternative for automated error detection in radiology reports, improving clinical workflow efficiency while ensuring patient data confidentiality. Further refinement could enhance their accuracy, contributing to better diagnosis and patient care., Competing Interests: Compliance with ethical standards. Guarantor: The scientific guarantor of this publication is Alexander Isaak, MD. Conflict of interest: The authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article. Statistics and biometry: No complex statistical methods were necessary for this paper. Informed consent: Informed consent was not required as no patient-identifying information was used. Ethical approval: After consultation, specific approval from the institutional review board was not required as no patient-identifying information was used. Study subjects or cohorts overlap: Not applicable. Methodology: Experimental, (© 2025. The Author(s).)

Background: Patients are increasingly using Generative Pre-trained Transformer 4 (GPT-4) to better understand their own radiology findings., Purpose: To evaluate the performance of GPT-4 in transforming cardiovascular magnetic resonance (CMR) reports into text that is comprehensible to medical laypersons., Methods: ChatGPT with GPT-4 architecture was used to generate three different explained versions of 20 various CMR reports (n = 60) using the same prompt: "Explain the radiology report in a language understandable to a medical layperson". Two cardiovascular radiologists evaluated understandability, factual correctness, completeness of relevant findings, and lack of potential harm, while 13 medical laypersons evaluated the understandability of the original and the GPT-4 reports on a Likert scale (1 "strongly disagree", 5 "strongly agree"). Readability was measured using the Automated Readability Index (ARI). Linear mixed-effects models (values given as median [interquartile range]) and intraclass correlation coefficient (ICC) were used for statistical analysis., Results: GPT-4 reports were generated on average in 52 s ± 13. GPT-4 reports achieved a lower ARI score (10 [9-12] vs 5 [4-6]; p < 0.001) and were subjectively easier to understand for laypersons than original reports (1 [1] vs 4 [4,5]; p < 0.001). Eighteen out of 20 (90%) standard CMR reports and 2/60 (3%) GPT-generated reports had an ARI score corresponding to the 8th grade level or higher. Radiologists' ratings of the GPT-4 reports reached high levels for correctness (5 [4, 5]), completeness (5 [5]), and lack of potential harm (5 [5]); with "strong agreement" for factual correctness in 94% (113/120) and completeness of relevant findings in 81% (97/120) of reports. Test-retest agreement for layperson understandability ratings between the three simplified reports generated from the same original report was substantial (ICC: 0.62; p < 0.001). Interrater agreement between radiologists was almost perfect for lack of potential harm (ICC: 0.93, p < 0.001) and moderate to substantial for completeness (ICC: 0.76, p < 0.001) and factual correctness (ICC: 0.55, p < 0.001)., Conclusion: GPT-4 can reliably transform complex CMR reports into more understandable, layperson-friendly language while largely maintaining factual correctness and completeness, and can thus help convey patient-relevant radiology information in an easy-to-understand manner., Competing Interests: Declaration of competing interests The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Alexander Isaak reports financial support was provided by BONFOR Research Commission of the Medical Faculty Bonn. The other authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper., (Copyright © 2024 The Author(s). Published by Elsevier Inc. All rights reserved.)

Objectives: To investigate whether proton density fat fraction (PDFF) measurements using a six-echo modified Dixon sequence can help to differentiate between benign and malignant vertebral bone marrow lesions., Methods: Sixty-six patients were prospectively enrolled in our study. In addition to conventional MRI at 3.0-Tesla including at least sagittal T2-weighted/spectral attenuated inversion recovery and T1-weighted sequences, all patients underwent a sagittal six-echo modified Dixon sequence of the spine. The mean PDFF was calculated using regions of interest and compared between vertebral lesions. A cut-off value of 6.40% in PDFF was determined by receiver operating characteristic curves and used to differentiate between malignant (< 6.40%) and benign (≥ 6.40%) vertebral lesions., Results: There were 77 benign and 44 malignant lesions. The PDFF of malignant lesions was statistically significant lower in comparison with benign lesions (p < 0.001) and normal vertebral bone marrow (p < 0.001). The areas under the curves (AUC) were 0.97 for differentiating benign from malignant lesions (p < 0.001) and 0.95 for differentiating acute vertebral fractures from malignant lesions (p < 0.001). This yielded a diagnostic accuracy of 96% in the differentiation of both benign lesions and acute vertebral fractures from malignancy., Conclusion: PDFF derived from six-echo modified Dixon allows for differentiation between benign and malignant vertebral lesions with a high diagnostic accuracy., Key Points: • Establishing a diagnosis of indeterminate vertebral lesions is a common clinical problem • Benign bone marrow processes may mimic the signal alterations observed in malignancy • PDFF differentiates between benign and malignant lesions with a high diagnostic accuracy • PDFF of non-neoplastic vertebral lesions is significantly higher than that of malignancy • PDFF from six-echo modified Dixon may help avoid potentially harmful bone biopsy.

Background Aortic stiffness and epicardial fat relate to cardiovascular risk. Their relationship with each other and their role with hypertension, diabetes mellitus (DM), and myocardial infarction (MI) can be evaluated by cardiac magnetic resonance (CMR). Purpose To explore an association between aortic stiffness and epicardial as well as paracardial fat volume (EFV and ParaFV, respectively) in hypertensive patients and to relate the results to the presence of DM and MI. Material and Methods A total of 156 hypertensive and 20 non-hypertensive participants were examined at 1.5 Tesla. A 2D-velocity-encoded sequence was acquired to assess aortic pulse wave velocity (PWV in m/s) as a measure of aortic stiffness. A 3D-Dixon sequence was used to determine EFV and ParaFV. Results PWV correlated with EFV (R = 0.474; P < 0.001), but not with ParaFV. Fat volumes (in mL/m 2 ) and PWV were lower in non-hypertensive controls compared to hypertensive patients. EFV and PWV were significantly higher in diabetic hypertensive patients without MI (n = 19; PWV: 10.4 ± 2.9; EFV: 92.5 ± 19.3) compared to hypertension-only patients (n = 84 [no DM or MI]; EFV: 64.8 ± 25.1, PWV: 9.0 ± 2.6; P < 0.05). Logistic regression analysis showed a significant association between the presence of a MI and a higher EFV ( P < 0.05), but not with PWV ( P = 0.060) or ParaFV ( P = 0.375). Conclusion A relationship between aortic stiffness and EFV was found in hypertensive patients. Both were increased in the presence of DM; however, only EFV was increased in the presence of MI. This may relate to the PWV lowering effect of the antihypertensive medication used by hypertensive patients and underscores the benefit of EFV assessment in this regard.

Objectives: To investigate the diagnostic value of cardiac magnetic resonance (CMR) feature-tracking (FT) myocardial strain analysis in patients with suspected acute myocarditis and its association with myocardial oedema., Methods: Forty-eight patients with suspected acute myocarditis and 35 control subjects underwent CMR. FT CMR analysis of systolic longitudinal (LS), circumferential (CS) and radial strain (RS) was performed. Additionally, the protocol allowed for the assessment of T1 and T2 relaxation times., Results: When compared with healthy controls, myocarditis patients demonstrated reduced LS, CS and RS values (LS: -19.5 ± 4.4% vs. -23.6 ± 3.1%, CS: -23.0 ± 5.8% vs. -27.4 ± 3.4%, RS: 28.9 ± 8.5% vs. 32.4 ± 7.4%; P < 0.05, respectively). LS (T1: r = 0.462, P < 0.001; T2: r = 0.436, P < 0.001) and CS (T1: r = 0.429, P < 0.001; T2: r = 0.467, P < 0.001) showed the strongest correlations with T1 and T2 relaxations times. Area under the curve of LS (0.79) was higher compared with those of CS (0.75; P = 0.478) and RS (0.62; P = 0.008)., Conclusions: FT CMR myocardial strain analysis might serve as a new tool for assessment of myocardial dysfunction in the diagnostic work-up of patients suspected of having acute myocarditis. Especially, LS and CS show a sufficient diagnostic performance and were most closely correlated with CMR parameters of myocardial oedema., Key Points: • Myocardial strain measures are considerably reduced in patients with suspected myocarditis. • Myocardial strain measures can sufficiently discriminate between diseased and healthy patients. • Myocardial strain measures show basic associations with the extent of myocardial oedema/inflammation.

Objectives: In clinical routine myocardial affection in amyloidosis is assessed by qualitative late gadolinium enhancement (LGE). Recent studies suggest that early cardiac involvement in amyloidosis may be overlooked by qualitative LGE assessment. The aim of this study was to assess possible early cardiac involvement in LGE negative AL patients by means of systolic and diastolic strain assessment and quantitative enhancement., Methods: 51 individuals (17 Patients with LGE positive light-chain amyloidosis (AL) (Group A), 17 Patients with LGE negative systemic AL (Group B), and 17 healthy controls (Group C)) were investigated. SSFP-Cine sequences were acquired in short axis slices as well as horizontal long axis views. Standard CMR parameters as well as Feature Tracking (FT) derived systolic and diastolic circumferential and longitudinal strain parameters were calculated. Additionally, contrast enhanced MRI images were analyzed to quantitatively determine the extent of enhancement., Results: AL patients with and without LGE both had significantly lower early diastolic strain rate (EDSR) values and peak systolic longitudinal strain (PSLS) values than healthy controls (EDSR: Group A 63.1±17.6; Group B: 74.46±11.8; Group C: 86.82±12.7; F(2.48)=10.7 p<0.001); (PSLS: Group A: -13.44±1.9%; Group B: -20.31±1.2%; Group C: -23.11±1.4%; F(2.48)=167.6; p<0.001). Analysis of quantitative LGE enhancement revealed increased enhancement in qualitative LGE negative AL patients compared to healthy controls (Group A: 19.6±8.9%; Group B: 8.2±3.9%; Group C: 2.4±1.3% F(2.48)=52.2; p<0.001)., Conclusion: CMR strain analysis detects early systolic and diastolic strain impairment in AL patients lacking qualitative LGE enhancement., (Copyright © 2016 Elsevier Ireland Ltd. All rights reserved.)

Background: Cardiac magnetic resonance (CMR) can detect inflammatory myocardial alterations in patients suspected of having acute myocarditis. There is limited information regarding the degree of normalization of CMR parameters during the course of the disease and the time window during which quantitative CMR should be most reasonably implemented for diagnostic work-up., Methods and Results: Twenty-four patients with suspected acute myocarditis and 45 control subjects underwent CMR. Initial CMR was performed 2.6±1.9 days after admission. Myocarditis patients underwent CMR follow-up after 2.4±0.6, 5.5±1.3, and 16.2±9.9 weeks. The CMR protocol included assessment of standard Lake Louise criteria, T1 relaxation times, extracellular volume fraction, and T2 relaxation times. Group differences between myocarditis patients and control subjects were highest in the acute stage of the disease (P<0.001 for all parameters). There was a significant and consistent decrease in all inflammatory CMR parameters over the course of the disease (P<0.01 for all parameters). Myocardial T1 and T2 relaxation times-indicative of myocardial edema-were the only single parameters showing significant differences between myocarditis patients and control subjects on 5.5±1.3-week follow-up (T1: 986.5±44.4 ms versus 965.1±28.1 ms, P=0.022; T2: 55.5±3.2 ms versus 52.6±2.6 ms; P=0.001)., Conclusions: In patients with acute myocarditis, CMR markers of myocardial inflammation demonstrated a rapid and continuous decrease over several follow-up examinations. CMR diagnosis of myocarditis should therefore be attempted at an early stage of the disease. Myocardial T1 and T2 relaxation times were the only parameters of active inflammation/edema that could discriminate between myocarditis patients and control subjects even at a convalescent stage of the disease., (© 2016 The Authors. Published on behalf of the American Heart Association, Inc., by Wiley Blackwell.)

Background: Today feature tracking (FT) is considered to be a robust assessment tool in cardiovascular magnetic resonance (CMR) for strain assessment. The FT algorithm is dependent on a high contrast between blood pool and myocardium. Extracellular contrast agents decrease blood-myocardial contrast in SSFP images and thus might affect FT results. However, in a routine CMR scan, SSFP-cine images including short axis views are partly acquired after contrast agent injection. The aim of this study was to investigate the effect of extracellular contrast agent (Gadobutrol) (CA) on the precision and reproducibility of the feature tracking algorithm., Methods: A total of 40 patient volunteers (mean age 51.2 ± 19 years; mean LVEF 61 ± 9 %) were scanned in supine position on a clinical 1.5 T MR scanner (Philips Ingenia). SSFP-cine images in midventricular short axis view (SA) as well as horizontal long axis view (HLA) were acquired before and 10-15 min after injection of a double dose Gadobutrol. FT derived systolic circumferential and longitudinal strain parameters were then calculated for pre- and post-contrast images., Results: FT derived midventricular peak systolic circumferential strain (PSCS) (-24.8 ± 6.4 % vs. -20.4 ± 6.3 %), apical PSCS (-28.67 ± 6.5 % vs. -24.06 ± 8.5 %), basal PSCS (-24.42 % ± 6.5 vs. -20.68 ± 7.1 %), peak systolic longitudinal strain (-19.57 ± 3.3 % vs. -17.24 ± 4.1 %), midventricular epicardial PSCS (-9.84 ± 3.4 % vs. -8.13 ± 3.4 %) , midventricular PSCS-rate (-1.52 ± 0.4 vs. -1.28 ± 0.5) and peak diastolic circumferential strain rate (1.4 ± 0.5 vs. 1.05 ± 0.5) were significantly reduced after CA application. Post CA strain assessment showed higher intra- and interobserver variability. Pre-CA: intraobserver: mean 0.21, Limits of agreement (LoA) -2.8 and 3.2; interobserver: mean 0.64, LoA -2.8 and 4.1. Post-CA: intraobserver: mean -0.11, LoA -5.1 to 4.9; interobserver: mean 4.93 LoA 2.4 to 12.2., Conclusion: The FT algorithm is dependent on a high contrast between blood and myocardium. Post CA strain results are significantly lower and less reproducible than pre-CA strain results.

Purpose: To investigate whether cardiac magnetic resonance (cardiac MR)-based feature tracking (FT) may be used for robust and rapid quantification of dyssynchrony by measurement of the septal to lateral delay (SLD)., Materials and Methods: Healthy volunteers (n = 18) and patients with mechanical dyssynchrony (n = 17) were investigated. Velocity encoding cardiac MR (VENC) and steady-state free precession (SSFP)-cine sequences were acquired in identical horizontal long axis (HLA) positions using a 1.5T MR scanner. Using FT and VENC cardiac MR software, myocardial velocity curves were calculated for the basal segment of the septal and lateral wall. Based on the quantity of dyssynchrony, the patients were classified into three subgroups (minimal, intermediate, extensive). SLD and patient classification were compared and intra- as well as interobserver variability assessed., Results: VENC and FT SLD measurements showed strong correlation (r = 0.94) and good agreement (mean 1.33 msec; limits of agreement [LoA] -2.96 to 5.63). Dyssynchrony subclassification based on FT was identical to VENC in 83% of the cases. While FT correctly classified all healthy subjects, three patients with mechanical dyssynchrony were misclassified. Compared to VENC, FT showed higher intra- and interobserver variability. VENC: intraobserver: mean 2.5 msec, LoA -17.5 to 22.5; interobserver: mean 1.5 msec, LoA -17.2 to 21.9. FT: intraobserver: mean 2.1 msec, LoA 27.6 to 31.8; interobserver: mean 2.4 msec LoA -31.4 to 34.5., Conclusion: Cardiac MR-based FT analysis may be used for rapid appraisal of left ventricle cardiac dyssynchrony from SSFP images. However, FT results are less accurate and reproducible compared to VENC-based assessment of SLD., (© 2015 Wiley Periodicals, Inc.)

Aim: Cardiac magnetic resonance (CMR) can visualize inflammatory tissue changes in acute myocarditis. Several quantitative image-derived parameters have been described to enhance the diagnostic value of CMR, but no direct comparison of these techniques is available., Methods and Results: A total of 34 patients with suspected acute myocarditis and 50 control subjects underwent CMR. CMR protocol included quantitative assessment of T1 relaxation times using modified Look-Locker inversion recovery (MOLLI) and shortened MOLLI (ShMOLLI) acquisition schemes, extracellular volume fraction (ECV), T2 relaxation times, and longitudinal strain. Established Lake-Louise criteria (LLC) consisting of T2-weighted signal intensity ratio (T2-ratio), early gadolinium enhancement ratio (EGEr), and late gadolinium enhancement (LGE) were assessed. Receiver operating characteristics analysis was performed to compare diagnostic performance. Areas under the curve of native T1 (MOLLI: 0.95; ShMOLLI: 0.92) and T2 relaxation times (0.92) were higher compared with those of the other CMR parameters (T2-ratio: 0.71, EGEr: 0.71, LGE: 0.87, LLC: 0.90, ECV MOLLI: 0.77, ECV ShMOLLI: 0.80, longitudinal strain: 0.83). Combined with LGE, each native mapping technique outperformed the diagnostic performance of LLC (P < 0.01, respectively). A combination of native parameters (T1, T2, and longitudinal strain) significantly increased the diagnostic performance of CMR compared with LLC without need of contrast media application (0.99 vs. 0.90; P = 0.008)., Conclusion: In patients suspected of having acute myocarditis, diagnostic performance of CMR can be improved by implementation of quantitative CMR parameters. Especially, native mapping techniques have the potential to replace current LLC. CLINICALTRIALS., Gov Number: NCT02299856., (Published on behalf of the European Society of Cardiology. All rights reserved. © The Author 2015. For permissions please email: journals.permissions@oup.com.)

There is growing evidence that pericardial and epicardial fat volume (PFV, EFV) are associated with cardiovascular risk. We evaluated a novel method for accurate measurement of PFV and EFV using a 3D-Dixon based cardiac magnetic resonance (CMR) approach. An electrocardiography triggered and respiratory navigator gated 3D-gradient echo pulse sequence was used for cardiac Dixon imaging. Based on this sequence, voxels predominantly containing fat were identified and added up for volumetry. After accuracy assessment in phantoms, consisting of muscle tissue and seven different fat samples (50-200 ml), the sequence was acquired in 34 healthy volunteers (22 male, BMI range 14-42 kg/m(2), age range 21-79 years) at 1.5 T. Analysis was performed independently by two readers who draw two 3D-regions of interest, one for EFV and one for PFV. Additionally, EFV and PFV were compared between overweighted and non-overweighted subjects. The phantom study showed an excellent agreement of measured and true fat volumes (maximum difference = 6 %, linear correlation coefficient R = 1.00). PFV over all volunteers was 158.0 ± 126.4 ml and EFV was 77.0 ± 55.3 ml. PFV and EFV were highly correlated (R = 0.96). Inter-reader agreement was good with a mean difference of 0.2 ± 5.6 and 4.5 ± 4.2 ml for PFV/EFV, (R > 0.99, each). EFV and PFV differed significantly between subjects with BMI > 25 kg/m(2) and BMI < 25 kg/m(2), n = 17 each (PFV 219.0 ± 151.8 vs. 96.9 ± 44.7 ml and EFV 102.3 ± 66.3 vs. 51.7 ± 23.6 ml, p < 0.001, each). The proposed 3D-Dixon based method allows accurate measurement of cardiac fat volumes. It provides a valuable tool for cardiovascular risk stratification by CMR.

To investigate the impact of a caffeine and taurine containing energy drink (ED) on myocardial contractility in healthy volunteers using cardiac MR and cardiac MR based strain analysis. 32 healthy volunteers (mean age 28 years) were investigated before and 1 h after consumption of a caffeine and taurine containing ED. For assessment of global cardiac functional parameters balanced SSFP-Cine imaging was performed, whereas CSPAMM tagging was used to evaluate global and regional myocardial strain. In addition, ten randomly chosen subjects were investigated once more using a caffeine only protocol to further evaluate the effect of caffeine solely. Heart rate and blood pressure were recorded throughout all studies. ED consumption led to a significant increase in peak systolic strain (PSS) and peak systolic strain rate (PSSR) 1 h after consumption (PSS: w/o ED -22.8 ± 2.1%; w ED -24.3 ± 2.4%, P = <0.0001 and PSSR: w/o ED -1.2 ± 0.1 1/s; w ED -1.3 ± 0.2 1/s, P = 0.0056), which was not observed in the caffeine only group. In contrast, global left ventricular function was unchanged (P = 0.2076). No significant changes of vital parameters and diastolic filling pattern were detected 1 h after ED consumption. Consumption of a caffeine and taurine containing ED results in a subtle, but significant increase of myocardial contractility 1 h after consumption.