European Radiology Experimental最新文献

Background: Breast magnetic resonance imaging (MRI) protocols often include T2-weighted fat-saturated (T2w-FS) sequences, which support tissue characterization but significantly increase scan time. This study aims to evaluate whether a 2D-U-Net neural network can generate virtual T2w-FS (VirtuT2w) images from routine multiparametric breast MRI images.

Methods: This IRB-approved, retrospective study included 914 breast MRI examinations from January 2017 to June 2020. The dataset was divided into training (n = 665), validation (n = 74), and test sets (n = 175). The U-Net was trained using different input protocols consisting of T1-weighted, diffusion-weighted, and dynamic contrast-enhanced sequences to generate VirtuT2. Quantitative metrics were used to evaluate the different input protocols. A qualitative assessment by two radiologists was used to evaluate the VirtuT2w images of the best input protocol.

Results: VirtuT2w images demonstrated the best quantitative metrics compared to original T2w-FS images for an input protocol using all of the available data. A high level of high-frequency error norm (0.87) indicated a strong blurring presence in the VirtuT2 images, which was also confirmed by qualitative reading. Radiologists correctly identified VirtuT2 images with at least 96% accuracy. Significant difference in diagnostic image quality was noted for both readers (p ≤ 0.015). Moderate inter-reader agreement was observed for edema detection on both T2w-FS images (κ = 0.49) and VirtuT2 images (κ = 0.44).

Conclusion: The 2D-U-Net generated virtual T2w-FS images similar to real T2w-FS images, though blurring remains a limitation. Investigation of other architectures and using larger datasets is necessary to improve potential future clinical applicability.

Relevance statement: Generating VirtuT2 images could potentially decrease the examination time of multiparametric breast MRI, but its quality needs to improve before introduction into a clinical setting.

Key points: Breast MRI T2w-fat-saturated (FS) images can be virtually generated using convolutional neural networks. Image blurring in virtual T2w-FS images currently limits their clinical applicability. Best quantitative performance could be achieved when using full dynamic-contrast-enhanced acquisition and DWI as input of the neural network.

Background: To compare liver image quality and lesion detection using an AI-augmented T1-weighted sequence on hepatobiliary-phase gadoxetate-enhanced magnetic resonance imaging (MRI).

Methods: Fifty patients undergoing gadoxetate-enhanced MRI were recruited. Two T1-weighted Dixon sequences were utilized: a 17-s breath-hold acquisition and an accelerated 12-s breath-hold acquisition (reduced phase resolution), both reconstructed using neural network (NN) and iterative denoising (ID), NN-alone, ID-alone, and the standard method. Contrast-to-noise ratio (CNR) was assessed quantitatively for all series (ANOVA). Two blinded radiologists independently analyzed three image sets: 17-s acquisition reconstructed with NN and ID (17-s NN + ID), 12-s acquisition reconstructed with NN and ID (12-s NN + ID), and 17-s acquisition with standard reconstruction (17-s standard). Overall image quality, qualitative CNR, lesion edge sharpness, vessel edge sharpness, and respiratory motion artifacts were scored (4-point Likert scale) and compared (Friedman test). Lesion detection was compared between 12-s NN + ID and 17-s standard reconstructions (Wilcoxon signed-rank test).

Results: Quantitative liver-to-portal vein CNR was significantly higher for 17-s NN + ID than 17-s standard or 17-s NN-alone images (p = 0.001). Scores for overall image quality, qualitative CNR, vessel edge sharpness, and lesion edge sharpness were significantly higher for 17-s NN + ID and 12-s NN + ID than standard reconstruction (p < 0.001); there was no significant difference between 17-s and 12-s NN + ID. There was no significant difference in respiratory motion artifacts and number of lesions or diameter of the smallest detected lesion using 12-s NN + ID or 17-s standard reconstruction.

Conclusion: AI-augmented reconstructions can improve image quality while reducing breath-hold duration in T1-weighted hepatobiliary-phase gadoxetate-enhanced MRI, without compromising lesion detection.

Relevance statement: AI-augmented reconstruction of T1-weighted MRI improves image quality and lesion detection in hepatobiliary phase liver imaging, reducing breath-hold duration without compromising clinical lesion detection.

Key points: Liver-to-portal vein CNR was significantly higher for 17-s NN + ID. AI-augmented reconstructions scored higher for image quality, contrast-to-noise, vessel-edge, and lesion-edge sharpness. No significant difference in lesion detection between 12-s NN + ID and 17-s standard reconstructions.

Background: We compared photon-counting detector computed tomography (PCD-CT) polyenergetic images, PCD-CT virtual monoenergetic images (VMI), and energy-integrating detector computed tomography (EID-CT) polyenergetic images regarding bone visualization and metal artifacts in patients with titanium wrist prostheses.

Methods: After ethical approval, 15 patients were examined with PCD-CT and EID-CT. Polyenergetic images were reconstructed, as well as 130-keV VMI for PCD-CT. Five radiologists evaluated bone visualization, interpretability at metal-bone interface and metal artifacts using a 7-point ordinal scale. Streak artifacts and artifacts at the bone-metal interface were quantitatively assessed. Differences between image setups were analyzed using Friedman test and one-way ANOVA with post hoc tests.

Results: Bone visualization was superior in PCD-CT polyenergetic images (median rating 6, range 3-7) compared with VMI (5, 3-7; p < 0.001) and EID-CT (5, 3-7; p = 0.018). Streak artifacts were more pronounced with PCD-CT polyenergetic images (4, 3-6) compared with EID-CT (5, 4-6; p = 0.003) and PCD-CT VMI (5, 3-7; p = 0.002), with quantitative results showing least streak artifacts in PCD-CT VMI, followed by EID-CT and PCD-CT polyenergetic images (50 ± 7%, 70 ± 6%, and 79 ± 5%, respectively; p < 0.001). Interpretability at bone-metal interface was better with PCD-CT polyenergetic images (5, 2-7; p = 0.045) and EID-CT (5, 3-6; p = 0.018) compared with PCD-CT VMI (4, 2-6), without quantitative differences.

Conclusion: Streak artifacts from titanium wrist prostheses were reduced using 130-keV PCD-CT VMI, while bone visualization was highest using PCD-CT polyenergetic images.

Relevance statement: In patients with wrist implants, photon-counting detector CT allows for effective metal artifact reduction using virtual monoenergetic images and improved bone visualization using polyenergetic images. As polyenergetic images and VMI have different advantages, access to both image setups may benefit diagnostic evaluation.

Key points: Virtual monoenergetic images (VMI) presented a substantial reduction of metal streak artifacts. Polyenergetic images exhibited better image quality for bone imaging compared with VMI. A combination of image reconstructions should be preferred depending on the diagnostic task.

Background: T2-weighted images are a critical component of prostate magnetic resonance imaging (MRI), and it would be useful to automatically assess image quality (IQ) on a patient-specific basis without radiologist oversight.

Methods: This retrospective study comprised 1,412 axial T2-weighted prostate scans. Four experienced uroradiologists graded IQ using a 0-to-3 scale (0 = uninterpretable; 1 = marginally interpretable; 2 = adequately diagnostic; 3 = more than adequately diagnostic), binarized into nondiagnostic (IQ0 or IQ1), requiring rescanning, and diagnostic (IQ2 or IQ3), not requiring rescanning. The deep learning (DL) model was trained on 1,006 scans; 203 other scans were used for validation of multiple convolutional neural networks; the remaining 203 exams were used as a test set. 3D-DenseNet_169 was chosen among 11 models based on multiple evaluation criteria. The rescan predictions were compared to the number of rescans performed on a subset of 174 exams.

Results: The model accurately predicts radiologist IQ scores (Cohen κ = 0.658), similar to the human inter-rater reliability (κ = 0.688-0.791). The model also predicts rescanning necessity similarly to radiologists: model κ = 0.537; reviewer κ = 0.577-0.703. The rescan model prediction area under the curve was 0.867.

Conclusion: The DL model showed a strong ability to differentiate diagnostic from nondiagnostic axial T2-weighted prostate images, accurately mimicking expert radiologists' IQ scores. Using the model, the clinical unnecessary rescan rate could be reduced from over 50% to less than 30%.

Relevance statement: DL assessment of T2-weighted prostate MRI scans can accurately assess IQ, determining the need to repeat inadequate scans as well as avoiding repeat scans of those with adequate diagnostic quality, resulting in reduced unnecessary rescanning.

Key points: Artificial intelligence assessment of prostate MRI T2-weighted image quality can improve exam time management. The model showed over 75% accuracy in assessing prostate MRI T2-weighted image quality. Expert radiologists have a substantial agreement in evaluating prostate MRI T2-weighted image quality.

Background: Phantoms play a critical role in mammography quality control (QC) by providing standardized conditions for evaluating image quality (IQ) metrics. However, inter-phantom variability may affect the reliability of these metrics, especially for inter-system comparisons. The aim of this study was to quantify the intra- and inter-phantom variability of IQ metrics using a set of theoretically identical phantoms.

Methods: Twenty-four TORMAS phantoms were imaged ten times each using a mammography unit under standardized high-dose conditions. Images were analyzed using automated software to extract 64 IQ metrics, including contrast-to-noise ratio (CNR) as well as modulation transfer function (MTF)-related and other metrics. Outliers were identified and excluded. Variability was assessed by calculating intra- and inter-phantom variances and coefficients of variation (COVs). The relative contributions of intra- and inter-phantom variability to total variability were also determined.

Results: Two defective phantoms were excluded. Analysis of 64 IQ metrics across 22 phantoms showed higher inter-phantom variability compared to intra-phantom variability. Mean intra- and inter-phantom COVs were 6.9% and 15.1% for the 34 CNR metrics, 4.8% and 5.4% for the 5 MTF-related metrics, 0.14% and 0.75% for the 10 contrast metrics, 4.9% and 14.8% for the 15 noise metrics, respectively. Inter-phantom variability contributed 84.2% to total variability, highlighting its dominance.

Conclusion: Inter-phantom variability significantly affects IQ metrics, emphasizing the importance of using the same phantom for inter-system comparisons to avoid confounding results. Conversely, phantoms are well-suited for assessing system reproducibility over time, focus on inter-system variability while consistently using a single phantom.

Relevance statement: This study highlights the significant impact of inter-phantom variability on image quality assessment, emphasizing the importance of using the same phantom for benchmarking imaging systems. These findings are crucial for optimizing quality control protocols and ensuring reliable, reproducible evaluations.

Key points: Inter-phantom variability exceeded intra-phantom variability across all image quality metrics of digital mammography. Subtle details showed higher total variability compared to more distinct features. Modulation transfer function metrics exhibited comparable intra- and inter-phantom variability, highlighting positioning sensitivity. Inter-phantom variability contributes 84% to total variability, impacting imaging system comparisons. Using the same phantom ensures reliability in imaging system performance evaluations.

Background: Early identification of treatment failure can effectively improve the success rate of antituberculosis treatment. This study aimed to construct a predictive model using radiomics based on cavity and cavity periphery to monitor the early treatment efficacy in multidrug-resistant tuberculosis (MDR-TB).

Methods: We retrospectively collected data on 350 MDR-TB patients who underwent pretreatment chest computed tomography (CT) and received longer regimens from two hospitals. They were subdivided into training (252 patients from hospital 1) and testing (98 patients from hospital 2) cohorts. According to at least two consecutive sputum culture results within the early sixth months of treatment, patients were divided into high-risk and low-risk groups. Radiomics models were established based on cavity and periphery with a range of 2, 4, 6, 8, and 10 mm. A combined model fused radiomics features of cavity with the best-performing peripheral regions. The performance of these models was evaluated by the receiver operating characteristic area under the curve (AUC) and clinical decision curve analysis.

Results: The cavity model achieved AUCs of 0.858 and 0.809 in the training and testing cohort, respectively. The radiomics model based on 4 mm peripheral region showed superior performance compared to other surrounding areas with AUCs of 0.884 and 0.869 in the two cohorts. The AUCs of the combined model were 0.936 and 0.885 in the two cohorts.

Conclusion: CT radiomics analysis integrating cavity and cavity periphery provided value in identifying MDR-TB patients at high risk of treatment failure. The optimal periphery extent was 4 mm.

Relevance statement: The cavity periphery also contains therapy-related information. Radiomics model based on cavity and 4 mm periphery is an effective adjunct to monitor early treatment efficacy for MDR-TB patients that can guide clinical decision.

Key points: A combined CT radiomics model integrating cavity with periphery can effectively monitor treatment response. A periphery of 4 mm showed superior performance compared to other peripheral smaller or greater extent. This study provided a surrogate for identifying the high risk of treatment failure in multidrug-resistant tuberculosis patients.

Background: Accurate assessment of treatment response to neoadjuvant systemic therapy (NAST) in breast cancer is important prior to surgery. We aimed at evaluating the feasibility of thoracic photon-counting detector computed tomography (PCCT) in assessing treatment response in breast cancers following NAST.

Methods: We retrospectively included patients with newly diagnosed breast cancer who received contrast-enhanced thoracic PCCT in prone position before and after NAST. Three experienced radiologists measured tumor size, tumor area, iodine uptake within tumors, number of suspicious breast lesions and of suspicious axillary lymph nodes before and after NAST. We compared the initial tumor size to contrast-enhanced magnetic resonance imaging (MRI), the residual tumor size after NAST to histopathology.

Results: Eighteen PCCT exams in nine patients aged 58 ± 14 years (mean ± standard deviation) were analyzed. After NAST, PCCT correctly identified a reduction in tumor burden in 9 of 9 cases and a complete response in 2 of 2 cases, with a significant reduction in tumor size, area, T-stage, number of suspicious breast lesions and of suspicious lymph nodes (p < 0.001 for all) as well as reduction in cutaneous infiltration (p = 0.010). Mean and maximum iodine uptake showed a nonsignificant reduction in cases with residual tumor after NAST (p = 0.092 and 0.363).

Conclusion: These preliminary findings suggest that thoracic PCCT can accurately detect local changes in breast cancer after NAST.

Relevance statement: Thoracic PCCT offers promising potential for accurately assessing breast cancer response to NAST.

Trial registration: German Clinical Trials Register DRKS00028997.

Key points: Prone thoracic contrast-enhanced photon-counting detector computed tomography (PCCT) can accurately detect reductions in tumor size, area, and T-stage. Prone PCCT can identify a decrease in the number of suspicious axillary lymph nodes. This technique shows promising results in identifying breast cancer response to neoadjuvant systemic therapy (NAST).

Background: Dark-field chest radiography is sensitive to the lung alveolar structure. We evaluated the change of dark-field signal between inspiration and expiration.

Methods: From 2018 to 2020, patients who underwent chest computed tomography (CT) were prospectively enrolled, excluding those with any lung condition besides emphysema visible on CT. Participants were imaged in both inspiration and expiration with a prototype dark-field chest radiography system. We calculated the total dark-field signal ∑DF and the dark-field coefficient ϵ, assumed to be proportional to the total number of alveoli and the alveolar density, respectively.

Results: Eighty-eight subjects, aged 64 years ± 11 (mean ± standard deviation), 55 males, were enrolled. Dark-field signal in the lung projection appeared higher in expiration compared to inspiration. Over all participants, ∑DF was higher in inspiration (1.6 × 10^-2 ± 0.4 × 10^-2 m²) compared to expiration (1.5 × 10^-2 ± 0.4 m²) (p < 0.001), with its expiration-to-inspiration not ratio being different for any emphysema subgroup. The dark-field coefficient ϵ was lower in inspiration (2.3 ± 0.6 m^-1) compared to expiration (3.1 ± 1.1 m^-1) (p < 0.001) over all participants. The dark-field coefficient in inspiration and expiration, as well as their ratio, was lower for at least moderate emphysema when compared to the control group (e.g., ϵ = 2.5 ± 1.0 m^-1 for moderate emphysema in expiration versus ϵ = 3.6 ± 0.7 m^-1 for participants without emphysema (p = 0.003).

Conclusion: The dark-field signal depends on the breathing state. Differences between breathing states are influenced by emphysema severity.

Relevance statement: The patient's breathing state influences the dark-field chest radiograph, potentially impacting its diagnostic value.

Key points: Signal characteristics in dark-field chest radiography change between inspiration and expiration. The total dark-field signal decreases slightly from inspiration to expiration, while the dark-field coefficient increases substantially. The ratio of the total dark-field signal between expiration and inspiration is independent of emphysema severity, whereas the ratio of the dark-field coefficient depends on emphysema severity.

Background: We investigated inflammation-induced changes in femoral hematopoietic bone marrow using advanced magnetic resonance imaging (MRI) techniques, including T2-weighted imaging, scalar T2 mapping, and machine learning-enhanced T2 distribution analysis to improve the detection of bone marrow microstructural alterations. Findings were correlated with histological markers and systemic inflammation.

Methods: Using a 9.4-T magnet, T2-weighted and multislice multiecho sequences were applied to evaluate bone marrow in female C57BL/6J mice divided into three groups: (1) controls; (2) lipopolysaccharide-induced acute inflammation (LPS); and (3) streptozotocin (STZ)- and LPS-induced diabetic inflammation (STZ + LPS). T2 relaxation times and their distributions with scalar mapping and model-informed machine learning (MIML) were analyzed. Correlations with histological iron levels and blood neutrophil counts were assessed.

Results: T2-weighted imaging showed a reduced signal-to-noise ratio in inflamed bone marrow (p = 0.034). Scalar T2 mapping identified decreased T2 relaxation times (p = 0.042), moderately correlating with neutrophil counts (ρ = 0.027) and iron levels (ρ = 0.016). MIML-enhanced T2 distribution analysis exhibited superior sensitivity than scalar T2 mapping, revealing significant reductions in the first T2 distribution peak (p = 0.0025), which strongly correlated with neutrophil counts (ρ = 0.0016) and iron sequestration (ρ = 0.0002). Histology confirmed elevated iron deposits in inflamed marrow, aligning with systemic inflammation.

Conclusion: Combining T2-weighted imaging, scalar T2 mapping, and MIML-enhanced T2 distribution analysis offers complementary insights into inflammation-induced bone marrow remodeling. T2 distribution analysis emerged as a more sensitive tool for detecting microstructural changes, such as iron sequestration, supporting its potential as a noninvasive biomarker for diagnosing and monitoring inflammatory diseases.

Relevance statement: This study highlights the potential of advanced MRI T2 analysis and machine learning methods for noninvasive detection of inflammation-induced microstructural changes in bone marrow, offering promising diagnostic tools for inflammatory diseases.

Key points: This study investigated inflammation-induced changes in bone marrow with T2 MRI and MIML. MIML outperformed quantitative scalar T2 analysis, increasingly detecting inflammation and iron sequestration in the hematopoietic bone marrow. T2 MRI with MIML analysis could aid in the early diagnosis and management of inflammatory diseases.

Over the past decades, computed tomography (CT) imaging has profited from various technical innovations. Besides improvements such as higher temporal and spatial resolutions, lower radiation dose, and the introduction of dual- and multi-energy imaging, the development and recent clinical introduction of photon-counting detector CT (PCD-CT) represents a milestone with the potential to substantially change clinical CT imaging and expand its indications. This thematic series of European Radiology Experimental comprises a collection of original research papers and review articles demonstrating the benefits and challenges of this cutting-edge technology. The thematic series includes a wide range of relevant topics spanning from initial clinical experiences using PCD-CT to original research papers covering potential applications in various body regions.