The large language model Generative Pretrained Transformer 4th-generation model (GPT-4; OpenAI) allows for the creation of summaries and graphic representations of the course of disease from the reports of various MRI examinations in patients with glioblastomas.

To investigate the potential of combining Compressed Sensing (CS) and a newly developed AI-based super resolution reconstruction prototype consisting of a series of convolutional neural networks (CNN) for a complete five-minute 2D knee MRI protocol. In this prospective study, 20 volunteers were examined using a 3T-MRI-scanner (Ingenia Elition X, Philips). Similar to clinical practice, the protocol consists of a fat-saturated 2D-proton-density-sequence in coronal, sagittal and transversal orientation as well as a sagittal T1-weighted sequence. The sequences were acquired with two different resolutions (standard and low resolution) and the raw data reconstructed with two different reconstruction algorithms: a conventional Compressed SENSE (CS) and a new CNN-based algorithm for denoising and subsequently to interpolate and therewith increase the sharpness of the image (CS-SuperRes). Subjective image quality was evaluated by two blinded radiologists reviewing 8 criteria on a 5-point Likert scale and signal-to-noise ratio calculated as an objective parameter. The protocol reconstructed with CS-SuperRes received higher ratings than the time-equivalent CS reconstructions, statistically significant especially for low resolution acquisitions (e.g., overall image impression: 4.3 ± 0.4 vs. 3.4 ± 0.4, p < 0.05). CS-SuperRes reconstructions for the low resolution acquisition were comparable to traditional CS reconstructions with standard resolution for all parameters, achieving a scan time reduction from 11:01 min to 4:46 min (57 %) for the complete protocol (e.g. overall image impression: 4.3 ± 0.4 vs. 4.0 ± 0.5, p < 0.05). The newly-developed AI-based reconstruction algorithm CS-SuperRes allows to reduce scan time by 57% while maintaining unchanged image quality compared to the conventional CS reconstruction.

Objectives: This study evaluated GPT-4’s accuracy in MRI sequence selection based on radiology request forms (RRFs), comparing its performance to radiology residents. Materials and methods: This retrospective study included 100 RRFs across four subspecialties (cardiac imaging, neuroradiology, musculoskeletal, and oncology). GPT-4 and two radiology residents (R1: 2 years, R2: 5 years MRI experience) selected sequences based on each patient’s medical history and clinical questions. Considering imaging society guidelines, five board-certified specialized radiologists assessed protocols based on completeness, quality, and utility in consensus, using 5-point Likert scales. Clinical applicability was rated binarily by the institution’s lead radiographer. Results: GPT-4 achieved median scores of 3 (1–5) for completeness, 4 (1–5) for quality, and 4 (1–5) for utility, comparable to R1 (3 (1–5), 4 (1–5), 4 (1–5); each p > 0.05) but inferior to R2 (4 (1–5), 5 (1-5); p < 0.01, respectively, and 5 (1–5); p < 0.001). Subspecialty protocol quality varied: GPT-4 matched R1 (4 (2–4) vs. 4 (2–5), p = 0.20) and R2 (4 (2–5); p = 0.47) in cardiac imaging; showed no differences in neuroradiology (all 5 (1–5), p > 0.05); scored lower than R1 and R2 in musculoskeletal imaging (3 (2–5) vs. 4 (3–5); p < 0.01, and 5 (3–5); p < 0.001); and matched R1 (4 (1–5) vs. 2 (1–4), p = 0.12) as well as R2 (5 (2–5); p = 0.20) in oncology. GPT-4-based protocols were clinically applicable in 95% of cases, comparable to R1 (95%) and R2 (96%). Conclusion: GPT-4 generated MRI protocols with notable completeness, quality, utility, and clinical applicability, excelling in standardized subspecialties like cardiac and neuroradiology imaging while yielding lower accuracy in musculoskeletal examinations. Key Points: Question Long MRI acquisition times limit patient access, making accurate protocol selection crucial for efficient diagnostics, though it’s time-consuming and error-prone, especially for inexperienced residents. Findings GPT-4 generated MRI protocols of remarkable yet inconsistent quality, performing on par with an experienced resident in standardized fields, but moderately in musculoskeletal examinations. Clinical relevance The large language model can assist less experienced radiologists in determining detailed MRI protocols and counteract increasing workloads. The model could function as a semi-automatic tool, generating MRI protocols for radiologists’ confirmation, optimizing resource allocation, and improving diagnostics and cost-effectiveness. Graphical

This study aimed to compare the image quality and diagnostic accuracy of deep-learning-based image denoising reconstructions (DLIDs) to established iterative reconstructed algorithms in low-dose computed tomography (LDCT) of patients with suspected urolithiasis. LDCTs (CTDIvol, 2 mGy) of 76 patients (age: 40.3 ± 5.2 years, M/W: 51/25) with suspected urolithiasis were retrospectively included. Filtered-back projection (FBP), hybrid iterative and model-based iterative reconstruction (HIR/MBIR, respectively) were reconstructed. FBP images were processed using a Food and Drug Administration (FDA)-approved DLID. ROIs were placed in renal parenchyma, fat, muscle and urinary bladder. Signal- and contrast-to-noise ratios (SNR/CNR, respectively) were calculated. Two radiologists evaluated image quality on five-point Likert scales and urinary stones. The results showed a progressive decrease in image noise from FBP, HIR and DLID to MBIR with significant differences between each method (p < 0.05). SNR and CNR were comparable between MBIR and DLID, while it was significantly lower in HIR followed by FBP (e.g., SNR: 1.5 ± 0.3; 1.4 ± 0.4; 1.0 ± 0.3; 0.7 ± 0.2, p < 0.05). Subjective analysis confirmed best image quality in MBIR, followed by DLID and HIR, both being superior to FBP (p < 0.05). Diagnostic accuracy for urinary stone detection was best using MBIR (0.94), lowest using FBP (0.84) and comparable between DLID (0.90) and HIR (0.90). Stone size measurements were consistent between all reconstructions and showed excellent correlation (r² = 0.958–0.975). In conclusion, MBIR yielded the highest image quality and diagnostic accuracy, with DLID producing better results than HIR and FBP in image quality and matching HIR in diagnostic precision.

Background/Objectives: To evaluate the diagnostic accuracy of two visual large language models (vLLMs), GPT-4o (OpenAI) and Claude Sonnet 3.5 (Anthropic), for detecting brain metastases in routine MRI using combined imaging and textual input. Methods: This retrospective study included 31 patients with and 46 without brain metastases with underlying melanoma (n = 24), lung cancer (n = 23), breast cancer (n = 17), or renal cell carcinoma (n = 13). In total, 100 MRI examinations (50 with, 50 without metastases) were provided to both vLLMs using a single representative slice per sequence, together with clinical history and the referring question. The generated free-text reports were evaluated for detection accuracy, overdiagnosis, correct sequence recognition, anatomical localization, lesion laterality, and lesion size estimation. Results: Both vLLMs showed perfect sensitivity (100% for both) but very low specificity (GPT-4o: 8%, Sonnet 3.5: 4%; p = 0.625), resulting in low diagnostic accuracy (GPT-4o: 54%, Sonnet 3.5: 52%; p = 0.625). Sequence identification was highly accurate in both models, with GPT-4o performing significantly better (100% vs. 93%; p < 0.05). Identification of the anatomical brain region (70% vs. 72%; p = 1.00) and lesion laterality (62% vs. 76%; p = 0.189) was comparable. Both models hallucinated additional lesions in 12% of cases. Lesion size measurements showed no significant differences between the models or in comparison with the radiologist. Conclusions: GPT-4o and Claude Sonnet 3.5 can generate radiological reports and detect brain metastases with excellent sensitivity, but their very low specificity, frequent hallucinations, and limited spatial reliability currently preclude clinical application. Future work should address how the balance between visual and textual input influences diagnostic behavior in vLLMs.

To evaluate the feasibility of combining Compressed SENSE (CS) with a newly developed deep learning-based algorithm (CS-AI) using a Convolutional Neural Network to accelerate balanced steady-state free precession (bSSFP)-sequences for cardiac magnetic resonance imaging (MRI). 30 healthy volunteers were examined prospectively with a 3 T MRI scanner. We acquired CINE bSSFP sequences for short axis (SA, multi-breath-hold) and four-chamber (4CH)-view of the heart. For each sequence, four different CS accelerations and CS-AI reconstructions with three different denoising parameters, CS-AI medium, CS-AI strong, and CS-AI complete, were used. Cardiac left ventricular (LV) function (i.e., ejection fraction, end-diastolic volume, end-systolic volume, and LV mass) was analyzed using the SA sequences in every CS factor and each AI level. Two readers, blinded to the acceleration and denoising levels, evaluated all sequences regarding image quality and artifacts using a 5-point Likert scale. Friedman and Dunn's multiple comparison tests were used for qualitative evaluation, ANOVA and Tukey Kramer test for quantitative metrics. Scan time could be decreased up to 57 % for the SA-Sequences and up to 56 % for the 4CH-Sequences compared to the clinically established sequences consisting of SA-CS3 and 4CH-CS2,5 (SA-CS3: 112 s vs. SA-CS6: 48 s; 4CH-CS2,5: 9 s vs. 4CH-CS5: 4 s, p < 0.001). LV-functional analysis was not compromised by using accelerated MRI sequences combined with CS-AI reconstructions (all p > 0.05). The image quality loss and artifact increase accompanying increasing acceleration levels could be entirely compensated by CS-AI post-processing, with the best results for image quality using the combination of the highest CS factor with strong AI (SA-CINE: Coef.:1.31, 95 %CI:1.05-1.58; 4CH-CINE: Coef.:1.18, 95 %CI:1.05-1.58; both p < 0.001), and with complete AI regarding the artifact score (SA-CINE: Coef.:1.33, 95 %CI:1.06-1.60; 4CH-CINE: Coef.:1.31, 95 %CI:0.86-1.77; both p < 0.001). Combining CS sequences with AI-based image reconstruction for denoising significantly decreases scan time in cardiac imaging while upholding LV functional analysis accuracy and delivering stable outcomes for image quality and artifact reduction. This integration presents a promising advancement in cardiac MRI, promising improved efficiency without compromising diagnostic quality.