Radiology advances最新文献

Background: Choroid plexus (ChP) has gained attention as a potential biomarker in neurodegenerative diseases, yet its segmentation remains challenging. Gadolinium-based contrast-enhanced MRI (CE-MRI) is the reference standard, as non-contrast MRI images lack sufficient contrast. However, gadolinium deposition, risk of nephrogenic system fibrosis in renally impaired patients, and patient discomfort limit its repeated administration.

Purpose: To develop deep learning-based synthetic-contrast-enhanced MRI (SynCE-MRI) using T1-weighted images to improve ChP visualization and evaluate its ability to detect morphological changes in Parkinsonian syndromes.

Materials and methods: This retrospective study included 265 (mean age = 65.7 ± 7.00 years, males/females: 120/145) consecutive patients in the internal cohort (174 with Parkinson's disease [PD], 46 with essential tremor, and 45 with atypical Parkinsonian disorder [APD]) who underwent T1W and CE-MRI at 3T from Asan Medical Center (June 2021-December 2023), and an external cohort of 58 (mean age = 60.7 ± 7.8 years, males/females: 40/18) patients (29/29 PD/APD) from Pusan National University, Yangsan Hospital (April 2011-December 2014). Nested-UNet was used for SynCE-MRI synthesis from T1W images. The 3D-UNet ChP segmentation model was trained by CE-MRI and tested using SynCE-MRI. Kruskal-Wallis and Bonferroni-corrected Mann-Whitney U tests assessed image synthesis, segmentation, and ChP morphometry (P < .05).

Results: SynCE-MRI achieved high-fidelity images with peak signal-to-noise ratio (PSNR) 35.37 ± 1.32 and structural similarity index measure (SSIM) 0.970 ±0.0054. Segmentation accuracy for SynCE-MRI (dice score = 0.803 ± 0.029, 95% CI: 0.797-0.810) significantly outperformed manual (dice score = 0.59 ± 0.057, 95% CI: 0.578-0.603; P < .001) and automated (dice score = 0.489 ± 0.049, 95% CI: 0.479-0.500; P < .001) T1W-based segmentations. SynCE-MRI-based ChP volumes closely matched CE-MRI (mean absolute-volume difference [MAVD] = 6.7%; ICC = 0.88, 95% CI: 0.82-0.92). SynCE-MRI revealed significantly larger ChP volumes in APD versus PD using internal cohort (APD: 2.69 ± 0.39 mL, 95% CI: 2.54-2.84 vs PD: 2.43 ± 0.47 mL, 95% CI: 2.25-2.52; P = .04) and external cohort (APD: 2.81 ± 0.48 mL, 95% CI: 2.60-3.02 vs PD: 2.52 ± 0.45 mL, 95% CI: 2.36-2.69; P = .03).

Conclusion: SynCE-MRI accurately replicates CE-MRI for ChP imaging and morphometry, outperforms T1W imaging in segmentation, and detects ChP enlargement in APD versus PD across internal and external cohorts, consistent with CE-MRI findings.

Background: Deep learning-based pancreas segmentation in CT has advanced rapidly yet remains evaluated primarily with mean overlap metrics that fail to capture robustness-defined as the proportion of cases reaching human-level performance. Models performing well on mean Dice or surface metrics can still fail unpredictably across scanners or anatomies. Because early detection and quantitative biomarkers rely on consistent segmentation, robustness is critical for clinical deployment.

Purpose: To systematically evaluate the robustness of deep learning models for pancreas segmentation relative to human readers and to investigate an active learning strategy to improve reliability.

Materials and methods: We retrospectively assembled 903 venous-phase CT scans from patients with presumed normal-appearing pancreases and without known pancreatic disease (2005-2023), split into 803 for training/validation and 100 healthy test cases. Each test case had 4 independent human segmentations. Inter-reader variability on this healthy-only test set defined the empirical human distribution, providing an upper-bound estimate of robustness. We introduced a Fractional Threshold (FT) metric, measuring the proportion of model predictions exceeding the minimum human performance. Robustness was assessed across models trained from scratch, fine-tuned, or pretrained, including both normal and abnormal cases. An active learning approach identified high-uncertainty predictions for human revision. Statistical comparisons were performed using the Wilcoxon signed-rank and proportions Z-tests.

Results: The best model, a 3-dimensional U-Net trained from scratch, achieved a Dice Similarity Coefficient (DSC) of 0.88 ± 0.04 and Normalized Surface Dice (NSD) of 0.77 ± 0.09, approaching human-level segmentation (DSC = 0.89 ± 0.03; NSD = 0.75 ± 0.07). However, FT for DSC and NSD remained lower than human performance in most cases, indicating persistent model variability. Human-in-the-loop revision of acquisition-flagged outliers increased FT to 0.99, with an average time of 1.54 minutes per case, corresponding to a 23-fold workload reduction.

Conclusion: Automated pancreas segmentation reduces workload but remains constrained by tail-case failures. Active learning enhances model reliability, bridging the gap between artificial intelligence and human-level performance.

Background: With chronic liver disease (CLD) rising globally, noninvasive methods are needed to stratify early, intermediate, and advanced CLD with and without clinically significant portal hypertension (CSPH).

Purpose: To analyze the combined diagnostic value of intravoxel incoherent motion (IVIM) and liver stiffness (LS) from magnetic resonance elastography (MRE) for CLD stage discrimination vs secondary single-parameter analyses.

Materials and methods: This retrospective cross-sectional study included 185 patients who underwent 3T liver MRI, including MRE and IVIM, between March 2016 and November 2023. Patients with CLD were grouped based on their liver fibrosis degree into early CLD (F0-F1; n = 21), intermediate CLD (F2; n = 19), advanced CLD (F3-F4, n = 20), and advanced CLD with CSPH (n = 22). CSPH was defined as splenomegaly (>120 mm) with thrombocytopenia (<100 × 10⁹/L), ascites, or portosystemic collaterals. Patients without CLD (n = 103) served as negative controls. IVIM parameters (tissue diffusivity D, perfusion fraction f, and pseudo-diffusion coefficient D*) and MRE LS were analyzed. Statistical analysis included the Kruskal-Wallis test and both univariate and multivariate regression.

Results: In total, 185 patients (median age: 55 years, interquartile range 25%-75%: 45-63 years; 94 men) were evaluated. All parameters differed significantly between all groups (P < .001). f and D* decreased with disease progression, while LS increased. D initially decreased in patients with CLD but increased in those with CSPH. Consequently, higher D-values indicated the presence of CSPH in advanced stages (odds ratio [OR] 1.09, 95% CI 1.03-1.17, P = .009). Elevated LS values showed strong associations with the presence of CLD (OR 5.01, CI 2.25-12.65, P < .001). Combining D and LS further improved diagnostic differentiation between disease stages, especially for differentiation between advanced CLD and advanced CLD with CSPH (OR 2.41, CI 1.33-5.44, P = .01).

Conclusion: IVIM and MRE are useful for characterizing CLD and CSPH. Combining D from IVIM with LS from MRE improves diagnostic accuracy compared to MRE alone.

Background: Noninvasive grading of liver inflammation in metabolic dysfunction-associated steatotic liver disease (MASLD) remains an unmet clinical need.

Purpose: To evaluate the diagnostic performance of multifrequency MR elastography (MMRE) for grading liver inflammation and diagnosing metabolic dysfunction-associated steatohepatitis (MASH).

Materials and methods: In this prospective, single-site study, participants underwent MMRE at 30, 40, and 60 Hz on a 3T system. Multifrequency dispersion coefficient (α), storage modulus ( $G\text{'}$ ), loss modulus ( $G"$ ), and magnitude of the shear modulus ( $\left|G^{*}\right|$ ) were computed. The reference standard was histopathological analysis of needle biopsy specimens. The MASLD activity score was computed to assign MASH status. Univariate and multivariable correlation and areas under the receiver operating characteristic curve (AUCs) were assessed.

Results: Of the 72 participants enrolled, 60 (12 healthy, 48 MASLD) were analyzable. Correlations of α, $G\text{'}$ , $G"$ , and $\left|G^{*}\right|$ with lobular inflammation grades were ρ = -0.62, P < .001; ρ = 0.46, P < .001; ρ = 0.39, P < .01; and ρ = 0.44, P < .001; and the corresponding correlations with ballooning ρ = -0.42, P < .001; ρ = 0.42, P < .001; ρ = 0.40, P < .001; and ρ = 0.41, P < .001. The correlation between α and lobular inflammation remained after adjusting for steatosis, ballooning, and fibrosis ( $\beta$ = -0.06, P < .001); and with ballooning after adjusting for steatosis, lobular inflammation, and fibrosis ( $\beta$ = -0.02, P < .01). AUCs of α were 0.85, 0.96, and 0.94, respectively, for distinguishing lobular inflammation grades 0 vs ≥1, ≤1 vs ≥2, and ≤2 vs 3; 0.84 and 0.78 for distinguishing ballooning grades 0 vs ≥1 and ≤1 vs 2; and 0.81 to 0.85 for diagnosing MASH.

Conclusion: The multifrequency dispersion coefficient α was associated with histologic inflammation and should be further evaluated with external validation as a possible clinical marker in MASLD.

Background: Benefits of ultrasound in breast cancer detection are often limited by the similar appearance of complicated cysts and solid hypoechoic masses with B-mode imaging, which can lead to false positive diagnoses.

Purpose: To evaluate the diagnostic performance of generalized contrast-to-noise ratio (gCNR) on short-lag spatial coherence (SLSC) ultrasound images as an objective tool to improve complicated cyst vs solid mass classification.

Materials and methods: For this secondary analysis of a prospective recruitment for the Advanced Ultrasound Signal Processing of Suspicious Breast Images (AUSPICIOUS) observational study (NCT07206888), women scheduled for ultrasound-guided procedures or follow-up of at least 1 breast mass were enrolled from March 2018 to October 2023. Raw ultrasound data were acquired with an Alpinion ECUBE12R research scanner, then post-processed with our custom software. The primary evaluation indicator was gCNR applied to SLSC images with regions of interest determined by 6 radiologists. Outcomes were compared to the same radiologists classifying mass contents as solid, fluid, mixed, or uncertain using B-mode images. The reference standard was determined by aspiration, pathology, or characterization of image features. Areas under receiver operating characteristic curves (AUCs) with 95% confidence intervals (CIs) and inter-reader agreement (Fleiss' κ) were assessed.

Results: Among 175 cases, 145 breast masses from 115 women (age: 52 ± 17 years) were analyzed, including 16 complicated cysts and 96 solid masses. The mean AUC for complicated cyst vs solid mass characterization was 0.96 (95% CI: 0.94, 0.97) with gCNR applied to SLSC images, relative to a mean lower-bound AUC of 0.67 (range: 0.54-0.76) with readings of B-mode images (P < .05). Inter-reader agreement improved from fair with B-mode (κ  =  0.40) to moderate with gCNR applied to SLSC images with a 0.76 threshold (κ  =  0.59, P < .00001).

Conclusion: Applying an objective gCNR metric to SLSC images improved the differentiation of complicated cysts from solid masses when compared to subjective readings of B-mode images.

Background: Atypical lipomatous tumors (ALTs) are aggressive fat cell tumors that are distinguished from benign lipomas (SL) mainly through histopathology. Biopsy is needed for suspicious cases but can miss malignancy, so complete surgical removal and examination are essential. MRI is used but often can't differentiate ALT from SL. We introduce a machine learning method for tumor classification.

Purpose: To characterize the classification performance of a Bayesian additive regression trees (BART) model, built from MR radiomic features, and compare it to the readings of a musculoskeletal radiologist in classifying atypical lipomatous tumors (ALTs) from simple lipomas.

Materials and methods: Retrospective data were collected from 5 medical institutions in North America, for a total of 437 patients; the mean age was 58 years ±12 years, with 248 men and 287 women. At each institution, at least T1-MRI images without contrast were collected from patients with suspected ALT prior to surgery. Histopathology was used as the reference standard. Radiomic features extracted from the MRI images were used to train the BART model and a random forest model for comparison of classification performance using a 10-fold cross-validation. Both models were compared with the classifications of an experienced (>10 years) musculoskeletal radiologist who scored the images on a 5-point scale.

Results: A cohort of 423 patients was included, and 1132 radiomic features were extracted from each MR study. The BART model had an accuracy, sensitivity, and specificity of 77.07% (72.76%-80.99%), 77.67% (71.36%-83.16%), and 76.50% (70.28%-81.97%), respectively, when utilizing all predictors and aggregating training and testing data from all the cohorts, approximating the human reader at 78.72% (74.51%-82.53%), 76.21% (69.80%-81.85%), and 81.11% (75.25%-86.09%), respectively. In the external validation, the average area under the curve (AUC) value across cohorts between the BART model and the human reader differed by 0.04 AUC points. From the receiver operating characteristic curve, the AUC was calculated to be 84.72% (81.00%-88.50%) and 84.74% (81.00%-88.50%) for the BART and human reader, respectively.

Conclusion: This study demonstrated that the BART model can distinguish ALT from lipoma with diagnostic performance comparable to an experienced human observer.

Background: Body composition metrics such as visceral fat volume, subcutaneous fat volume, and skeletal muscle volume are important predictors of cardiovascular disease, diabetes, and cancer prognosis.

Purpose: We explore the use of deep learning to estimate body composition metrics from chest radiographs and a small set of easily obtainable clinical variables.

Materials and methods: A retrospective cohort of patients with concurrent noncontrast abdominal CT's and frontal chest radiographs within 3 months of each other was selected. A multitask, multimodal, deep learning model using chest radiographs and clinical variables (age, sex at birth, height and weight extracted from electronic medical records) was trained to estimate the body composition metrics. Reference standard was body composition, including subcutaneous fat volume, measured on CT.

Results: Our final cohort consisted of 1118 patients (582 female and 538 male subjects) from 30 health systems across the United States with imaging performed from 2010 to 2024. The mean age at imaging was 67 years (SD: 17), mean height was 1.67 meters (SD: 0.2), and mean weight was 78 kg (SD: 20). Average values for visceral fat, subcutaneous fat, and skeletal muscle indices were 59.39 cm²/m² (SD: 39.26), 88.13 cm²/m² (SD: 58.52), and 44.81 cm²/m² (SD: 15.49). The best-performing model achieved a Pearson correlation of 0.85 (95% CI: 0.81-0.88) for subcutaneous fat volume, 0.76 (0.65-0.80) for visceral fat volume, and 0.58 (0.49-0.67) for skeletal muscle volume with the multimodal model outperforming unimodal models (P = .0001 for subcutaneous fat volume). Mean absolute errors of the best performing models for subcutaneous and visceral fat volumes were 1054 cm³/m² and 667 cm³/m², respectively.

Conclusion: We introduced a multimodal deep learning model leveraging chest radiographs to estimate body composition. Our model can facilitate large-scale studies by estimating body composition using a chest radiograph and commonly available clinical variables.

Background: Metabolic dysfunction-associated steatotic liver disease affects 1 in 3 people worldwide. Ultrasound shear wave elastography (SWE) in obese patients, the target population for testing, is hampered by beam attenuation, leading to unreliable liver fibrosis quantification.

Purpose: We assess the safety and efficacy of increased push mechanical index (IPMI) above U.S. Food and Drug Administration limits to improve SWE.

Materials and methods: This single-center prospective trial (July 2023-April 2024) (NCT05792423) enrolled healthy adults stratified by body mass index (BMI). Participants underwent conventional push pulse (mechanical index [MI] 1.4) and IPMI (MI 2.5) SWE (GE Healthcare LOGIQ E10) performed by 1 of 3 sonographers and serial liver function testing (LFT) before and up to 7 days after imaging. Liver injury was defined as increased serum alanine transaminase (ALT), aspartate aminotransferase (AST), or alkaline phosphatase (ALP) (non-inferiority margins: AST 7.5 U/L, ALT 12 U/L, ALP 17.5 U/L). Secondary endpoints included SWE variability and measurement number.

Results: Twenty-two analyzable participants (mean age 39.6 ± 16.3 years; 15 women) had normal BMI (6), overweight (6), class 1 obesity (7), and class 2 obesity (3). Conventional shear wave speed was 1.34 ± 0.21 m/s, and IPMI yielded 1.36 ± 0.20 m/s (velocities ≤ 1.34 m/s indicate minimal or no fibrosis). The mean [95% CI] LFT change from baseline to day 1 was 1) AST: -0.86 [-2.34, 0.61], P = .24, 2) ALT: 0.32 [-1.04, 1.68], P = .63, 3) ALP: 1.73 [-1.02, 4.47], P = .21. The upper 95% CI for all biomarkers met non-inferiority criteria. Mean IPMI interquartile range (IQR) to median ratio decreased 0.019 (29.2% relative reduction) (P = .01) with 0.68 [IQR: 0.0.75] (P = .058), fewer average attempts.

Conclusion: IPMI SWE in healthy volunteers did not cause injury and reduced measurement variability. IPMI SWE should be developed to improve examination quality and reliability in obese patients.