Privacy-ensuring, open-weights large language models showed great potential in the extraction of structured content from free-text radiology reports, facilitating the secondary use of clinical databases for applications in data-driven medicine.

Previous studies suggest an impact of body composition on outcome in melanoma patients. We aimed to determine the prognostic value of CT-based body composition assessment in patients receiving immune checkpoint inhibitor therapy for treatment of metastatic disease using a deep learning approach. One hundred seven patients with staging CT examinations prior to initiation of checkpoint inhibition between January 2013 and August 2019 were retrospectively evaluated. Using an automated deep learning-based body composition analysis pipeline, parameters for estimation of skeletal muscle mass (skeletal muscle index, SMI) and adipose tissue compartments (visceral adipose tissue index, VAI; subcutaneous adipose tissue index, SAI) were derived from staging CT. The cohort was binarized according to gender-specific median cut-off values. Patients below the median were defined as having low SMI, VAI, or SAI, respectively. The impact on outcome was assessed using the Kaplan–Meier method with log-rank tests. A multivariable logistic regression model was built to test the impact of body composition parameters on 3-year mortality. Patients with low SMI displayed significantly increased 1-year (25% versus 9%, p = 0.035), 2-year (32% versus 13%, p = 0.017), and 3-year mortality (38% versus 19%, p = 0.016). No significant differences with regard to adipose tissue compartments were observed (3-year mortality: VAI, p = 0.448; SAI, p = 0.731). On multivariable analysis, low SMI (hazard ratio (HR), 2.245; 95% confidence interval (CI), 1.005–5.017; p = 0.049), neutrophil-to-lymphocyte ratio (HR, 1.170; 95% CI, 1.076–1.273; p < 0.001), and Karnofsky index (HR, 0.965; 95% CI, 0.945–0.985; p = 0.001) remained as significant predictors of 3-year mortality. Lowered skeletal muscle index as an indicator of sarcopenia was associated with worse outcome in patients with metastatic melanoma receiving immune checkpoint inhibitor therapy.

Background: The acquisition of contrast‐enhanced T1 maps to calculate extracellular volume (ECV) requires contrast agent administration and is time consuming. This study investigates generative adversarial networks for contrast‐free, virtual extracellular volume (vECV) by generating virtual contrast‐enhanced T1 maps. Methods and Results: This retrospective study includes 2518 registered native and contrast‐enhanced T1 maps from 1000 patients who underwent cardiovascular magnetic resonance at 1.5 Tesla. Recent hematocrit values of 123 patients (hold‐out test) and 96 patients from a different institution (external evaluation) allowed for calculation of conventional ECV. A generative adversarial network was trained to generate virtual contrast‐enhanced T1 maps from native T1 maps for vECV creation. Mean and SD of the difference per patient (ΔECV) were calculated and compared by permutation of the 2‐sided t test with 10 000 resamples. For ECV and vECV, differences in area under the receiver operating characteristic curve (AUC) for discriminating hold‐out test patients with normal cardiovascular magnetic resonance versus myocarditis or amyloidosis were tested with Delong's test. ECV and vECV showed a high agreement in patients with myocarditis (ΔECV: hold‐out test, 2.0%±1.5%; external evaluation, 1.9%±1.7%) and normal cardiovascular magnetic resonance (ΔECV: hold‐out test, 1.9%±1.4%; external evaluation, 1.5%±1.2%), but variations in amyloidosis were higher (ΔECV: hold‐out test, 6.2%±6.0%; external evaluation, 15.5%±6.4%). In the hold‐out test, ECV and vECV had a comparable AUC for the diagnosis of myocarditis (ECV AUC, 0.77 versus vECV AUC, 0.76; P =0.76) and amyloidosis (ECV AUC, 0.99 versus vECV AUC, 0.96; P =0.52). Conclusions: Generation of vECV on the basis of native T1 maps is feasible. Multicenter training data are required to further enhance generalizability of vECV in amyloidosis.

Purpose: To determine if deep learning (DL) segmentation of total fetal body volume (TFBV) and total fetal lung volume (TFLV) in fetuses with congenital diaphragmatic hernia has comparable performance to manual segmentation. Materials and Methods: A total of 208 fetal MRI studies with congenital diaphragmatic hernia, acquired between August 2007 and September 2023, were retrospectively included. TFBV and TFLV were extracted from manual tissue segmentations in balanced gradient echo and single shot T2-weighted turbo spin echo sequences. MRI studies were split into training (n = 188) and hold-out test data (n = 20). Wilcoxon signed-rank test was used to compare manual and DL-based segmentations by 2 U-Nets. Manual and DL segmentation times were noted and compared using Student’s t-test. The observed/expected ratio of the total lung volume (O/E TLV) as a prognostic marker for postnatal survival was calculated. Outcome predictions of O/E TLV for postnatal death were assessed with univariate regression analysis. Results: Manual segmentation times were higher compared to DL segmentations (30 ± 7 minutes versus 0.25 ± 0.05 minutes, P < .001). Manual and DL-based TFBV were similar (1317 ± 498 mL versus 1306 ± 491 mL; P = .04; Dice score: 0.98 ± 0.01). TFLV (19.4 ± 11.5 mL versus 18.7 ± 12.4 mL; P = .11; Dice score: 0.84 ± 0.09) and O/E TLV (39.3 ± 18.1 mL versus 37.7 ± 19.1 mL, P = .13) were not significantly different. Postnatal mortality was negatively associated with higher manual O/E TLV (odds ratio: 0.97; 95% confidence interval [CI], 0.96–0.98; P < .001) and DL O/E TLV (odds ratio: 0.97; 95% CI, 0.96–0.98; P < .001). Conclusion: DL for body and lung segmentation in fetuses with congenital diaphragmatic hernia allows reliable and rapid calculations of the observed/expected ratio and equally predicts prognostic outcome.

Objectives: Recently, epicardial adipose tissue (EAT) assessed by CT was identified as an independent mortality predictor in patients with various cardiac diseases. Our goal was to develop a deep learning pipeline for robust automatic EAT assessment in CT. Methods: Contrast-enhanced ECG-gated cardiac and thoraco-abdominal spiral CT imaging from 1502 patients undergoing transcatheter aortic valve replacement (TAVR) was included. Slice selection at aortic valve (AV)-level and EAT segmentation were performed manually as ground truth. For slice extraction, two approaches were compared: A regression model with a 2D convolutional neural network (CNN) and a 3D CNN utilizing reinforcement learning (RL). Performance evaluation was based on mean absolute z-deviation to the manually selected AV-level (Δz). For tissue segmentation, a 2D U-Net was trained on single-slice images at AV-level and compared to the open-source body and organ analysis (BOA) framework using Dice score. Superior methods were selected for end-to-end evaluation, where mean absolute difference (MAD) of EAT area and tissue density were compared. 95% confidence intervals (CI) were assessed for all metrics. Results: Slice extraction using RL was slightly more precise (Δz: RL 1.8 mm (95% CI: [1.6, 2.0]), 2D CNN 2.0 mm (95% CI: [1.8, 2.3])). For EAT segmentation at AV-level, the 2D U-Net outperformed BOA significantly (Dice score: 2D U-Net 91.3% (95% CI: [90.7, 91.8]), BOA 85.6% (95% CI: [84.7, 86.5])). The end-to-end evaluation revealed high agreement between automatic and manual measurements of EAT (MAD area: 1.1 cm² (95% CI: [1.0, 1.3]), MAD density: 2.2 Hounsfield units (95% CI: [2.0, 2.5])). Conclusions: We propose a method for robust automatic EAT assessment in spiral CT scans enabling opportunistic evaluation in clinical routine. Critical relevance statement: Since inflammatory changes in epicardial adipose tissue (EAT) are associated with an increased risk of cardiac diseases, automated evaluation can serve as a basis for developing automated cardiac risk assessment tools, which are essential for efficient, large-scale assessment in opportunistic settings. Key Points: Deep learning methods for automatic assessment of epicardial adipose tissue (EAT) have great potential. A 2-step approach with slice extraction and tissue segmentation enables robust automated evaluation of EAT. End-to-end automation enables large-scale research on the value of EAT for outcome analysis. Graphical