European Radiology Experimental最新文献

Background: To compare the quality and clinical usefulness of large language model (LLM)-generated lumbar spine magnetic resonance imaging (MRI) reports with radiologist-written ones and assess whether medical professionals can distinguish between them.

Materials and methods: This retrospective observational single-center study was approved by the local ethics committee. A total of 125 lumbar spine MRI reports (104 human-written, 21 LLM-generated using ChatGPT-4o) were anonymized, randomized, and blindly evaluated by five medical professionals (one board-certified radiologist, two radiology residents, one general practitioner, one orthopedic surgeon), all with basic familiarity with LLM. Each report was scored on a five-point Likert scale for clinical relevance, clarity, completeness, diagnostic accuracy, and intelligibility, whereas general practitioner and orthopedic surgeon evaluated intelligibility only. Evaluators also classified each report as AI-generated or human-written. Accuracy was defined as the proportion of correctly classified reports in distinguishing LLM-generated from radiologist-written texts. Mann-Whitney U or Student's t-tests were used.

Results: Radiologists' reports consistently received higher median scores across all domains (p < 0.001). No differences were found in the description of the imaging technique (p > 0.175). No clinically false statements were identified in the LLM-generated reports. Identification accuracy varied widely among evaluators: Board-certified radiologist achieved 88.0% accuracy (sensitivity 66.7%, specificity 92.3%), Resident 1 65.6% (14.3%, 76.0%), Resident 2 94.4% (66.7%, 100%), orthopedic surgeon 78.4% (90.5%, 76.0%) and general practitioner 65.6% (81.0%, 62.5%).

Conclusion: Radiologist-written lumbar spine MRI reports outperform LLM-generated reports in quality and structure. However, some AI-generated reports were indistinguishable from human ones, particularly for non-specialized readers. LLMs may support radiologists in structured reporting and improve workflow efficiency, while maintaining diagnostic reliability.

Relevance statement: Large language models can draft lumbar spine MRI reports, but currently lack the quality and consistency of radiologist reports. With radiologist supervision, large language models may improve reporting efficiency while preserving diagnostic reliability and supporting clinical decision-making.

Key points: LLM-generated reports are clinically coherent and stylistically comparable to those written by expert radiologists. Radiologist-written reports scored significantly higher for clinical relevance, findings, and structure. LLM-generated reports were sometimes misclassified as human-written by clinicians.

Objective: New magnetic resonance imaging (MRI) gradient technology enables the acquisition of ultrahigh b-value diffusion-weighted imaging (DWI). We assessed its impact on image quality and Prostate Imaging Reporting and Data System (PI-RADS) scores in prostate MRI.

Materials and methods: Participants with cancer suspicion prospectively underwent 3-T prostate MRI (maximum gradient strength 200 mT/m). Sequences with b-values of 0/800, 1,500, 2,500, 3,500, and 4,500 s/mm² were acquired. Lesion conspicuity was rated from 1 (non-diagnostic) to 5 (excellent). Apparent signal-to-noise ratios (aSNR) and acquisition times were determined. Cumulative link mixed-effects models, repeated measures ANOVA, and Cohen/Fleiss κ statistics were used.

Results: A total of 107 participants, aged 67 ± 8 years (mean ± standard deviation), were included. Compared to DWI(b1500), the DWI(b2500), DWI(b3500), and DWI(b4500) acquisitions were worse regarding both lesion conspicuity (median score, 5 [interquartile interval 4-5] versus 4 [3-4] versus 2 [2-3] versus 2 [1-2], respectively; all p < 0.001) and aSNR (19.0 ± 7.5 versus 12.7 ± 4.8 versus 11.8 ± 4.1 versus 11.4 ± 2.6, respectively; all p < 0.001). Acquisition times increased from DWI(b1500) (107 ± 9 s) to DWI(b4500) (329 ± 26 s). Cohen κ for PI-RADS score agreement was good to moderate (DWI(b2500): 0.87 [confidence interval 0.81, 0.94]; DWI(b3500): 0.75 [0.65, 0.84]; DWI(4500): 0.61 [0.49, 0.72]).

Conclusion: Acquired ultrahigh gradient DWI sequences with ultrahigh b-values in prostate MRI had worse image quality than standard b-values, while PI-RADS agreement between DWI(b1500) and DWI(b2500) was good. However, diagnostic estimates for clinically significant prostate carcinoma remained limited due to a small biopsy sample size (50/107 patients).

Relevance statement: Ultrahigh b-value DWI showed no improved diagnostic performance in comparison to standard b-value DWI regarding the identification of potential prostate cancer. Ultrahigh b-value should not replace standard high b-values (1,500 s/mm²) for imaging workup of patients with suspicion for prostate cancer.

Key points: Acquired ultrahigh b-values (b2500-4500) using ultrahigh gradients of up to 140 T/m were utilized for prostate DWI. Both, overall image quality and diagnostic confidence decreased from good for DWI(b1500) to non-diagnostic for DWI(b4500). PI-RADS agreement between DWI(b1500) and DWI(b2500) was good, while it was only moderate between DWI(b1500) and DWI(b4500).

Objectives: Cold ischemia during kidney transplantation induces ischemia-reperfusion injury with endothelial dysfunction, capillary leak, and impaired perfusion. Its duration critically determines graft outcome. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) enables noninvasive assessment of renal microcirculation and may indicate ischemic injury. We evaluated the impact of ischemia duration on DCE-MRI-derived perfusion parameters in renal transplants in mice.

Materials and methods: Procedures were approved by the local institutional animal care and use committee. A total of 15 C57BL/6 mice underwent kidney transplantation and were assigned to a short or prolonged cold ischemia group. DCE-MRI was performed to assess renal perfusion. Imaging was conducted at a mean of 268 ± 30 days (mean ± standard deviation) after transplantation. Perfusion parameters were calculated using the Patlak model, which provides the plasma volume fraction (v_p), reflecting renal blood volume and perfusion, and the volume transfer constant (K^trans), characterizing the rate of contrast agent extravasation from capillaries into the extravascular extracellular space.

Results: Significant differences were observed in the K^trans parameter of transplanted kidneys between groups. The median K^trans (mL/100 mL/min) was significantly higher in the 16-h group (2.87, interquartile range 2.45-3.03) versus the 30-min group (0.91, 0.90-1.42; p = 0.008). Median v_p (mL/100 mL/min) was non-significantly lower in the 16-h group (21.89, 17.28-23.22) versus the 30-min group (29.02, 24.99-37.15; p = 0.151).

Conclusion: Cold ischemia with 16-h duration was associated with significantly higher K^trans values in kidney transplants, reflecting significantly increased vascular permeability. DCE-MRI provides a sensitive tool for detecting ischemia-induced microvascular dysfunction.

Relevance statement: Quantitative DCE-MRI detects microvascular injury after 16-h cold ischemia in kidney transplants in mice, supporting its potential as a noninvasive tool to assess graft integrity and guide interventions aimed at improving long-term transplant outcomes.

Key points: The duration of ischemia critically affects endothelial integrity and perfusion characteristics in a mouse kidney transplant model. Prolonged 16-h ischemia leads to increased vascular permeability, indicating more severe endothelial and microcirculatory injury in transplanted kidneys. DCE-MRI enables sensitive detection of subtle ischemia-related microvascular alterations, supporting its value for noninvasive graft assessment.

Objective: We investigated the transport of gadolinium-based contrast agent (GBCA) across the blood-brain barrier (BBB) along the perivascular spaces as part of the glymphatic drainage in patients with iatrogenic BBB disruption following digital subtraction angiography (DSA).

Materials and methods: A retrospective analysis was conducted on patients who underwent DSA for diagnosis and/or treatment of intracranial aneurysms and received a 3-T magnetic resonance imaging (MRI) within the following day. Exclusion criteria included states with a suggested impairment of BBB integrity, such as neurodegenerative diseases or suspected glymphatic impairment. BBB disruption was assessed using a pre- and post-contrast three-dimensional T1-weighted volume-isotropic turbo spin-echo sequence. Patterns of GBCA distributions were described. The localization of GBCA-extravasation was correlated with perivascular spaces visualized on the coregistered T2-weighted sequences. Fisher's exact test and logistic regression were used.

Results: Out of 43 patients, 30 (69.8%) exhibited visible BBB disruption. BBB disruption was significantly more often observed after therapeutic DSA (p = 0.004). GBCA-enhancement patterns indicated a localized pial enhancement in 96.7% of affected patients, with additional parenchymal enhancement along the perivascular spaces in 56.7%. Enhancement was predominantly located in the downstream territories of probed vessels, suggesting a potential association with glymphatic transport. An illustrative case with serial MRI examinations is presented, demonstrating time-dependent GBCA-enhancement patterns.

Conclusion: The study provides in vivo evidence of GBCA transport patterns following iatrogenic BBB disruption, which may correspond to parts of the proposed glymphatic pathways. Our results indicate a sequential progression of contrast enhancement, initially manifesting at the brain surface and subsequently extending along perivascular spaces to the subarachnoid space.

Relevance statement: Understanding BBB disruption and glymphatic transport with MRI imaging methods may improve neurovascular disease management.

Key points: BBB disruption post-DSA may facilitate GBCA transport via glymphatic pathways, offering novel and hypothesis-generating insights into brain clearance mechanisms. GBCA enhancement followed a chronological and spatial pattern, suggesting an organized cerebrospinal-interstitial exchange system relevant for brain clearance. Findings highlight potential implications for BBB integrity in neurovascular health with prospective implications for diagnostic imaging.

Objective: We compared a super-resolution deep learning image reconstruction (SR-DLR) algorithm with a normal-resolution (NR)-DLR algorithm according to radiation dose for abdominal computed tomography (CT).

Materials and methods: An image-quality phantom was scanned with an energy-integrating detectors CT unit at three volume CT dose index radiation dose levels (12.7, 5.9, and 3 mGy). Images were reconstructed using a 1,024² matrix for SR-DLR and a 512² matrix for NR-DLR, for three DLR levels (level-1, level-2, and level-3). Noise power spectrum (NPS) and task-based transfer function (TTF) for iodine and Solid Water^® inserts were computed; TTF values at 50% (f₅₀, mm^-1) were used to quantify spatial resolution. The detectability index (d') was computed for two simulated lesions.

Results: Noise magnitude values were lower with SR-DLR than with NR-DLR for level-2 (-27.6 ± 3.8%) and level-3 (-43.5 ± 1.4%), the opposite for level-1. Average NPS spatial frequency was higher with SR-DLR than with NR-DLR for all radiation dose levels for level-1 (55.9 ± 16.7%) and level-2 (20.1 ± 13.9%) and the opposite for level-3, except at 12.7 mGy. For both inserts, f₅₀ was higher with SR-DLR than with NR-DLR at each radiation dose and DLR level. For simulated lesions and all DLR levels, d' values were higher with SR-DLR than with NR-DLR (level-1, 6.0 ± 2.0%; level-2, 45.7 ± 5.0%; level-3, 75.2 ± 7.3%).

Conclusion: Compared to NR-DLR, SR-DLR improved spatial resolution and detectability of simulated abdominal lesions; image noise was reduced with SR-DLR only for level-2 and level-3, while image texture was better for level-1 and level-2.

Relevance statement: Super-resolution DLR with a 1,024² matrix size improved spatial resolution and detectability of simulated abdominal lesions compared to normal-resolution DLR. Validation in clinical settings is necessary before translation into routine practice.

Key points: The performance of a new deep learning super-resolution image reconstruction algorithm (SR-DLR) was compared to a normal-resolution (NR)-DLR algorithm using an image-quality phantom for an abdominal energy-integrating detector CT protocol. SR-DLR with a 1,024² matrix improved spatial resolution and detectability of simulated abdominal lesions compared to NR-DLR with a 512² matrix. Using SR-DLR, therefore, presents numerous prospects for improving abdominal CT images and a high potential for reducing the radiation doses.

Objective: Abdominal aortic aneurysm (AAA) remains a life-threatening condition with few large-animal disease models. We aimed to develop a fully endovascular porcine AAA model for radiology research, reducing surgical trauma and improving reproducibility versus laparotomy-based models.

Materials and methods: Fourteen female German Landrace swine (n = 14, 30-40 kg) underwent angiography-guided intervention. The animals' infrarenal aorta was dilated by ~30% via balloon catheter, then collagenase (6,000 IU), elastase (500 IU), and 25% calcium chloride (0.5 mL) were locally incubated to weaken the vessel wall. Eight animals were included in the study; group 1 (n = 4) was euthanized at 2 weeks, and group 2 (n = 4) at 4 weeks. Aortic diameter was measured weekly by ultrasound; ex vivo histology, immunofluorescence, and western blot assessed remodeling and inflammation.

Results: Progressive aneurysm expansion was observed, with diameters of 1.32 ± 0.08 cm (mean ± standard deviation) at 1 week post-intervention, 1.59 ± 0.06 cm at 2 weeks, 1.81 ± 0.10 cm at 3 weeks, and 1.94 ± 0.19 cm at 4 weeks (baseline: 0.74 ± 0.08 cm; p < 0.001). Experimental groups' macrophages increased (group 1, 15.12 ± 3.88%; group 2, 16.65 ± 5.27%) compared to control (0.66 ± 0.27%, p = 0.012 and p = 0.021, respectively). Vascular smooth muscle cells were reduced across interventional groups (45.97 ± 17.26% versus control 80.94 ± 14.26%, p = 0.005).

Conclusions: This porcine AAA model replicates human disease features with a fully endovascular workflow, offering a valuable platform for evaluation of novel imaging techniques and interventional therapies.

Relevance statement: This study presents a fully endovascular porcine model of abdominal aortic aneurysm for translational research in interventional radiology and imaging. By enabling aneurysm induction entirely through catheter-based techniques, the model could provide a clinically relevant platform for future evaluation of novel endovascular devices and intraluminal therapeutics.

Key points: This study established a fully endovascular, translational porcine model of abdominal aortic aneurysm. The model exhibited a significant mean aneurysmal dilation of about 161% at 4 weeks and 107% at 2 weeks. Serial ultrasound confirmed consistent aneurysm expansion and reproducible growth patterns in surviving animals. Ex vivo analyses demonstrated inflammation and extracellular-matrix damage, mirroring key features of human abdominal aortic aneurysm pathology. This fully catheter-based workflow provides a practical preclinical platform for evaluating imaging techniques and endovascular therapies.

Artificial intelligence (AI) and large language models (LLMs) are increasingly integrated into radiology, offering new possibilities for advanced imaging techniques, including cardiovascular magnetic resonance (CMR). This proof-of-concept study assessed four high-performing LLMs (Gemini 2.5 Pro, ChatGPT 4.1, DeepSeek V3, and Claude Opus 4) on their ability to generate CMR protocols for 140 hypothetical cardiac cases. AI-generated protocols were compared against a reference standard established by a consensus between two experienced cardiovascular radiologists, following the Society for Cardiovascular Magnetic Resonance (SCMR) recommendations. Descriptive statistics were used to quantify the concordance of LLM-generated sequences with the SCMR guidelines. Statistical agreement was measured using Cohen and Fleiss κ statistics. Gemini 2.5 Pro achieved the highest concordance, aligning with the SCMR guidelines in 71.5% of all evaluated scenarios. Overall, LLMs showed moderate agreement with the SCMR protocols, with Gemini 2.5 Pro again performing best (Cohen κ = 0.55). Agreement was substantial for mandatory CMR sequences (Fleiss κ ∈ [0.69, 0.74]) and predominantly fair for optional sequences. The tested LLMs demonstrate a potential to generate efficient and pathology-adapted CMR protocols. Under expert supervision, this capability could streamline the imaging workflow and help extend CMR to primary healthcare centers through protocol automation. RELEVANCE STATEMENT: The potential of Gemini 2.5 Pro, ChatGPT 4.1, DeepSeek V3, and Claude Opus 4 to suggest pathology-adapted CMR protocols could improve imaging throughput and help to expand access to advanced cardiac diagnostics in primary healthcare centers. KEY POINTS: The tested large language models show potential for generating CMR protocols. Substantial agreement on mandatory CMR sequences promises more efficient examinations. Automation of CMR protocols could help to improve access to this advanced technique outside major medical institutions.

Background: Chemical exchange saturation transfer (CEST)-magnetic resonance imaging (MRI), particularly amide proton transfer-weighted (APTw)-CEST and 2-deoxy-D-glucose-CEST, holds promise for noninvasive molecular breast cancer (BC) characterization. However, quantification remains challenging due to field inhomogeneities, overlapping exchange pools, and the limited robustness of conventional metrics such as the magnetization transfer ratio asymmetry (MTR_asym). This study evaluates four CEST postprocessing metrics-MTR_asym, Lorentzian amplitudes, MTR relaxation exchange (MTR_REX), and apparent exchange-dependent relaxation (AREX)-for their diagnostic performance in differentiating BC subtypes using endogenous APTw-CEST and exogenous 2-deoxy-D-glucose-CEST in a murine BC xenograft model of Luminal A, human epidermal growth factor receptor 2 (HER2)+, and triple-negative tumors.

Materials and methods: Metabolic CEST-MRI was performed in vitro on protein and 2-deoxy-D-glucose phantoms and in vivo in a murine BC model. Imaging was conducted at 9.4 T with 120 frequency offsets from +6 to -6 ppm. MTR_REX and AREX were derived via Lorentzian fitting using tailored five-pool models. Statistical comparisons across subtypes were performed per metric.

Results: In APTw-CEST, MTR_REX and AREX significantly distinguished Luminal A from HER2+ (p ≤ 0.027) and Luminal A from triple-negative (p ≤ 0.006) tumors. Lorentzian amplitudes differentiated Luminal A from triple-negative (p = 0.019), while MTR_asym showed no separation. In 2-deoxy-D-glucose-CEST, only AREX distinguished Luminal A from HER2+ tumors (p = 0.017).

Conclusion: Advanced metrics, particularly MTR_REX and AREX, improve metabolic CEST-MRI for BC subtyping in a murine preclinical model, while MTR_asym is inadequate for this purpose.

Relevance statement: Our findings underscore the importance of applying advanced postprocessing metrics to metabolic CEST-MRI for improved noninvasive BC characterization in a murine preclinical model.

Key points: Advanced multimetabolic APTw-CEST and 2-deoxy-D-glucose-CEST postprocessing metrics allowed adequate preclinical murine BC subtyping. AREX showed potential for 2-deoxy-D-glucose-CEST in tumor characterization; however, APTw-CEST remains superior. MTR_asym failed to distinguish between tumor subtypes in CEST-MRI.