European Radiology Experimental最新文献

Background: We investigated the influence of different kernel types and sharpness levels on in vitro and in vivo coronary stenosis quantification in high-pitch photon-counting detector coronary CT angiography (PCD-CCTA).

Materials and methods: Coronary stenoses were evaluated in a phantom containing two stenosis grades (25% and 50%), and in a retrospective cohort of 30 patients who underwent high-pitch PCD-CCTA. Scans were reconstructed as virtual monoenergetic images at 55 keV using three different kernels (Br, Bv, and Qr) and four sharpness levels (36, 40, 44, and 48). Percent diameter stenosis (PDS) values were compared. In vitro measurements were additionally compared with the stenosis reference value. Two readers independently assessed the in vivo measurements.

Results: In vitro, PDS values of all stenoses showed no difference among various kernel types and sharpness levels (p ≥ 0.412). However, PDS measurements using kernel Bv40 showed the smallest cumulative deviation from the ground truth. In vivo, a total of 53 stenoses were identified in 30 patients, aged 63 ± 13 years (mean ± standard deviation), 8/30 (27%) females. There was no significant difference in PDS measurements among reconstructions, either when analyzed per stenosis or stratified by different plaque types (p = 1.000). Bv kernels showed higher interobserver reliability (intraclass correlation coefficient: Bv 0.91; Qr 0.88; Br 0.85).

Conclusion: With comparable diagnostic accuracy, different kernel types and sharpness levels can be used in high-pitch PCD-CCTA. Due to the in vivo advantage in interobserver reliability and the in vitro observed lowest cumulative deviation from ground truth, reconstruction with kernel Bv40 should be preferred.

Relevance statement: For image reconstruction in PCD-CCTA with high-pitch mode, kernel Bv40 should be considered to obtain the best diagnostic performance and reliability of stenosis quantification.

Key points: High-pitch PCD-CCTA images can be reconstructed with different kernels. Reconstructions with different kernels showed comparable accuracy on coronary stenosis quantification. In vitro, Bv40 reconstructions showed superior measurement accuracy to the reference. In vivo, reconstructions with the Bv kernel had the highest interobserver reliability. Reconstruction with kernel Bv40 should be considered in high-pitch PCD-CCTA.

Background: Periodontitis is characterized by the inflammatory destruction of tooth-supporting alveolar bone. Dental magnetic resonance imaging (MRI) using dynamic contrast-enhanced perfusion can potentially detect vascular inflammatory responses. This study aims to assess the feasibility of perfusion dental MRI and characterize periodontal lesions with perfusion profiles.

Materials and methods: In this prospective study, 19 patients with severe periodontitis underwent pretreatment 3-T dental MRI with T2-weighted, high-resolution dynamic contrast-enhanced T1-weighted perfusion protocol, and contrast-enhanced T1-weighted fat-suppressed sequences as well as cone-beam computed tomography (CBCT). Periodontal bone lesions were segmented semiautomatically using a multistep threshold-based algorithm, guided by T1-weighted contrast enhancement, T2-weighted hyperintensity, as well as CBCT-based bone loss. Volumetric analyses and clinical data were compared with perfusion parameters.

Results: In all 95 assessed periodontal lesions, perfusion parameter elevations were significantly different when compared to normal distant bone (p < 0.001 to 0.026). Moreover, structurally normal-appearing bone adjacent to T2-hyperintense/T1-contrast-enhancing signal alterations exhibited increased permeability (p = 0.036-006) but showed no significant change in blood flow (p = 0.270) compared to bone control areas. Lesions with bleeding showed higher vascular permeability and blood flow markers than lesions without bleeding (p = 0.004-0.006). Additionally, lesions with excessive edema and areas of bone loss exhibited significantly elevated permeability and blood flow parameters (p = 0.001-0.028).

Conclusion: Perfusion dental MRI for periodontal lesion assessment is feasible. Permeability/perfusion parameters elevations are related to clinical signs of inflammation and CBCT-based bone loss, with the potential for detecting early inflammatory responses.

Relevance statement: Perfusion dental MRI effectively characterizes periodontal disease by detecting inflammation-related vascular changes beyond structural imaging on CBCT and conventional MR, offering potential for improved diagnosis, monitoring, and treatment evaluation. Longitudinal studies are needed.

Key points: Perfusion dental MRI detects increased blood flow and vascular permeability in periodontal lesions. Increased permeability in adjacent bone suggests early inflammatory changes before structural loss. Dental MRI perfusion metrics could aid early lesion detection and monitoring of periodontitis.

Background: Manufacturer-defined AI thresholds for chest x-ray (CXR) often lack customization options. Threshold optimization strategies utilizing users' clinical real-world data along with pathology-enriched validation data may better address subgroup-specific and user-specific needs.

Materials and methods: A pathology-enriched dataset (study cohort, 563 (CXRs)) with pleural effusions, consolidations, pneumothoraces, nodules, and unremarkable findings was analysed by an AI system and six reference radiologists. The same AI model was applied to a routine dataset (clinical cohort, 15,786 consecutive routine CXRs). Iterative receiver operating characteristic analysis linked achievable sensitivities (study cohort) to resulting AI alert rates in clinical routine inpatient or outpatient subgroups. "Optimized" thresholds (OTs) were defined by a 1% sensitivity increase leading to more than a 1% rise in AI alert rates. Threshold comparisons (OTs versus AI vendor's default thresholds (AIDT) versus Youden's thresholds) were based on 400 clinical cohort cases with expert radiologists' reference.

Results: AIDTs, OTs, and Youden's thresholds varied across scenarios, with OTs differing based on tailoring for inpatient or outpatient CXRs. AIDT lowering most reasonably improved sensitivity for pleural effusion, with increases from 46.8% (AIDT) to 87.2% (OT) for outpatients and from 76.3% (AIDT) to 93.5% (OT) for inpatients; similar trends appeared for consolidations. Conversely, regarding inpatient nodule detection, increasing the threshold improved accuracy from 69.5% (AIDT) to 82.5% (OT) without compromising sensitivity. Graphical analysis supports threshold selection by illustrating estimated sensitivities and clinical routine AI alert rates.

Conclusion: An innovative, subgroup-specific AI threshold optimization is proposed, automatically implemented and transferable to other AI algorithms and varying clinical subgroup settings.

Relevance statement: Individually customizing thresholds tailored to specific medical experts' needs and patient subgroup characteristics is promising and may enhance diagnostic accuracy and the clinical acceptance of diagnostic AI algorithms.

Key points: Customizing AI thresholds individually addresses specific user/patient subgroup needs. The presented approach utilizes pathology-enriched and real-world subgroup data for optimization. Potential is shown by comparing individualized thresholds with vendor defaults. Distinct thresholds for in- and outpatient CXR AI analysis may improve perception. The automated pipeline methodology is transferable to other AI models or subgroups.

Background: Spectral photon-counting computed tomography (SPCCT) outperformed dual-energy computed tomography (DECT) for coronary artery stenosis assessment. However, data about myocardial perfusion imaging (MPI) is lacking. This feasibility study aimed to evaluate and compare the diagnostic performance of SPCCT and DECT for rest MPI in patients with hemodynamically significant coronary stenoses, using invasive coronary angiography (ICA) and invasive fractional flow reserve (FFR) as reference standards.

Materials and methods: Eighteen very-high-risk patients with hemodynamically significant coronary stenoses at ICA underwent both dual-layer DECT and SPCCT coronary CT within three days. The sensitivity, specificity, and accuracy of MPI in detecting myocardial hypoperfusion were assessed. Quantitative attenuation differences between normal and hypoperfused myocardial segments were compared for both modalities. Interobserver variability was assessed with a weighted kappa analysis.

Results: SPCCT demonstrated comparable overall performance to DECT for MPI, with an overall sensitivity, specificity, and accuracy of 73.3%, 79.2%, and 76.9%, respectively, versus 73.3%, 75%, and 74.4% for DECT. SPCCT outperformed DECT in the left anterior descending artery territory, achieving a sensitivity of 87.5%, specificity of 100%, and accuracy of 90%, versus 62.5%, 50%, and 60% for DECT. For each CT system, attenuation analysis revealed differences between normal and hypoperfused segments, with mean differences of 17.9 HU for DECT and 15.8 HU for SPCCT (p < 0.05). Inter-reader agreement was higher for SPCCT (κ = 0.86) compared to DECT (κ = 0.62).

Conclusion: SPCCT and DECT provided similar diagnostic performance for rest MPI in a very-high-risk patient cohort, demonstrating comparable effectiveness in detecting the effects of hemodynamically significant coronary stenosis.

Relevance statement: Hemodynamically significant stenosis in very-high-risk patients results in myocardial hypoperfused areas at rest that can be detected equally well with dual-layer CT and spectral photon counting CT, albeit with better reproducibility for the latter.

Key points: SPCCT and DECT showed comparable performance for MPI at rest in very-high-risk patients. The differences between normal and hypoperfused segments were of 17 HU and 16 HU on conventional images for DECT and SPCCT. SPCCT showed higher interobserver agreement compared to DECT, suggesting improved reproducibility.

Background: Effective doctor-patient communication (DPC) skills are critical competencies in residency training. This study evaluated the efficacy of a smart glass (SG)-based communication skills training curriculum for radiology residents in China.

Materials and methods: This quasi-experimental study with a one-group pretest-posttest design involved 18 radiology residents in an 8-week SG-based DPC simulation training. Supervisors used the SEGUE scale, while residents and four standardised patients (SPs) trained by the Guangdong Institute of Simulation Medicine self-assessed satisfaction with a Likert scale. Analysis compared pre- and post-training scores (before, immediately after, and 6 months post-programme). Post-training SG experiences were assessed via surveys.

Results: Significant improvements were observed in SEGUE scale scores immediately and 6 months post-programme compared with pre-programme scores (17.06 ± 3.67 and 17.72 ± 3.12 versus 10.94 ± 2.88, respectively, p < 0.001). Similarly, Likert scores for SPs and residents showed significant increases both immediately and 6 months post-programme compared with initial scores (3.50 ± 0.51 and 3.67 ± 0.68 versus 2.39 ± 0.61, p < 0.001 for both, and 3.28 ± 0.46 and 3.55 ± 0.78 versus 2.66 ± 0.84, p = 0.037 and 0.008, respectively). Post-training, the Likert consistency between SPs and residents was 0.73 (p = 0.005). Of 18 participants, 16 (89%) reported that SG provided useful feedback, and 16 (89%) recognised the value of SG in developing DPC skills.

Conclusion: The SG-based simulation training programme significantly enhanced and sustained DPC skills among radiology residents.

Relevance statement: Smart glasses provide an innovative tool for recording standardised patient encounters, offering a perspective for analysing and evaluating residents' interpersonal communication skills and nonverbal behaviours.

Key points: Smart glasses enhance doctor-patient communication skills in radiology residents. Simulation training with smart glasses showed improvement in skills. Smart glasses offer a perspective for standardised patient encounters. They facilitate better analysis of residents' interpersonal and nonverbal communication.

Background: Annotating new classes in MRI images is time-consuming. Refining presegmented structures can accelerate this process. Many target classes lacking in MRI are supported by computed tomography (CT) models, but translating MRI to synthetic CT images is challenging. We demonstrate that CT segmentation models can create accurate MRI presegmentations, with or without image inversion.

Materials and methods: We retrospectively investigated the performance of two CT-trained models on MRI images: a general multiclass model (TotalSegmentator); and a specialized renal tumor model trained in-house. Both models were applied to 100 T1-weighted (T1w) and 100 T2-weighted fat-saturated (T2wfs) MRI sequences from 100 patients (50 male). Segmentation quality was evaluated on both raw and intensity-inverted sequences using Dice similarity coefficients (DSC), with reference annotations comprising manual kidney tumor annotations and automatically generated segmentations for 24 abdominal structures.

Results: Segmentation quality varied by MRI sequence and anatomical structure. Both models accurately segmented kidneys in T2wfs sequences without preprocessing (TotalSegmentator DSC 0.60), but TotalSegmentator failed to segment blood vessels and muscles. In T1w sequences, intensity inversion significantly improved TotalSegmentator performance, increasing the mean DSC across 24 structures from 0.04 to 0.56 (p < 0.001). Kidney tumor segmentation demonstrated poor performance in T2wfs sequences regardless of preprocessing. In T1w sequences, inversion improved tumor segmentation DSC from 0.04 to 0.42 (p < 0.001).

Conclusion: CT-trained models can generalize to MRI when supported by image augmentation. Inversion preprocessing enabled segmentation of renal cell carcinoma in T1w MRI using a CT-trained model. CT models might be transferable to the MRI domain.

Relevance statement: CT-trained artificial intelligence models can be adapted for MRI segmentation using simple preprocessing, potentially reducing manual annotation efforts and accelerating the development of AI-assisted tools for MRI analysis in research and future clinical practice.

Key points: CT segmentation models can create presegmentations for many structures in MRI scans. T1w MRI scans require an additional inversion step before segmenting with a CT model. Results were consistent for a large multiclass model (i.e., TotalSegmentator) and a smaller model for renal cell carcinoma.

Background: Automated segmentation methods may potentially increase the reliability and applicability of radiomics in skeletal oncology. Our aim was to propose a deep learning-based method for automated segmentation of atypical cartilaginous tumor (ACT) and grade II chondrosarcoma (CS2) of long bones on magnetic resonance imaging (MRI).

Materials and methods: This institutional review board-approved retrospective study included 164 patients with surgically treated and histology-proven cartilaginous tumors at two tertiary bone tumor centers. The first cohort consisted of 99 MRI scans from center 1 (79 ACT, 20 CS2). The second cohort consisted of 65 MRI scans from center 2 (45 ACT, 20 CS2). Supervised Edge-Attention Guidance segmentation Network (SEAGNET) architecture was employed for automated image segmentation on T1-weighted images, using manual segmentations drawn by musculoskeletal radiologists as the ground truth. In the first cohort, a total of 1,037 slices containing the tumor out of 99 patients were split into 70% training, 15% validation, and 15% internal test sets, respectively, and used for model tuning. The second cohort was used for independent external testing.

Results: In the first cohort, Dice Score (DS) and Intersection over Union (IoU) per patient were 0.782 ± 0.148 and 0.663 ± 0.175, and 0.748 ± 0.191 and 0.630 ± 0.210 in the validation and internal test sets, respectively. DS and IoU per slice were 0.742 ± 0.273 and 0.646 ± 0.266, and 0.752 ± 0.256 and 0.656 ± 0.261 in the validation and internal test sets, respectively. In the independent external test dataset, the model achieved DS of 0.828 ± 0.175 and IoU of 0.706 ± 0.180.

Conclusion: Deep learning proved excellent for automated segmentation of central cartilage tumors on MRI.

Relevance statement: A deep learning model based on SEAGNET architecture achieved excellent performance for automated segmentation of cartilage tumors of long bones on MRI and may be beneficial, given the increasing detection rate of these lesions in clinical practice.

Key points: Automated segmentation may potentially increase the reliability and applicability of radiomics-based models. A deep learning architecture was proposed for automated segmentation of appendicular cartilage tumors on MRI. Deep learning proved excellent with a mean Dice Score of 0.828 in the external test cohort.

Background: Spectral computed tomography (CT) late-enhancement (LE) acquisitions can help detect myocarditis. An arterial acquisition is often performed for coronary artery analysis. However, little is known about the appearance of myocarditis on the arterial phase. We investigated the appearance of myocarditis on arterial acquisitions of cardiac spectral CT, and its relationship to LE and edema.

Materials and methods: Forty-seven cardiac spectral CTs performed in patients with magnetic resonance imaging (MRI)-confirmed myocarditis were retrospectively assessed. Three myocardial attenuation/enhancement patterns were visually identified and segmented on both arterial and LE acquisitions: hypodense-arterial + normal-LE (HypoArt-NorLE); normal-arterial + hyperdense-LE (NorArt-HyperLE); and hypodense-arterial + hyperdense-late (HypoArt-HyperLE). Characteristics of conventional and spectral images were calculated for all patterns and for remote myocardium. Values of HypoArt-HyperLE lesions were compared in the groups with and without edema on MRI, as assessed with T2 mapping (available for 25 patients).

Results: We found 173 lesions, 46 (26%) HypoArt-NorLE, 54 (31%) NorArt-HyperLE, and 73 (42%) HypoArt-HyperLE. On the arterial phase, HypoArt-HyperLE were more hypodense (p < 0.001) and had less iodine (0.23 mg/mL less; p < 0.001) than RM. On LE, both HypoArt-HyperLE and NorArt-HyperLE were more hyperdense and contained more iodine than the remote myocardium (all p < 0.001). HypoArt-HyperLE lesions were more hypodense and contained less iodine on the arterial phase in patients with edema on MRI as compared to those without (all p < 0.001).

Conclusion: Most myocarditis lesions detectable with spectral CT are visible on both arterial and LE acquisitions. These lesions appeared to be more pronounced on the arterial phase in patients with edema on MRI.

Relevance statement: Spectral CT arterial acquisition performed for the differential diagnosis of acute myocardial pathologies in many cases can depict myocarditis lesions as epicardial hypodense areas, most likely related to the presence of edema.

Key points: Data from spectral CT shows that most myocarditis lesions appear as hypodense on the arterial phase, matching the epicardial LE zones. A minority of myocarditis lesions appear as epicardial LE areas without anomalies of attenuation on the arterial phase. Hypodense myocardial areas are correlated to the presence of edema on MRI, suggesting they are due to the same phenomenon.

Background: Bone marrow (BM) lesion differentiation remains challenging, and quantitative magnetic resonance imaging (MRI) may enhance accuracy over conventional methods. We evaluated the diagnostic value and inter-reader reliability of Dixon-based signal drop (%drop) and fat fraction percentage (%fat) as adjuncts to existing protocols.

Materials and methods: In this prospective two-center study, 172 patients with BM signal abnormalities underwent standardized 1.5-T MRI protocols, including Dixon sequences. Two musculoskeletal radiologists independently evaluated images and performed quantitative measurements of %drop and %fat. Final diagnoses were established through histopathology (n = 96) or imaging follow-up (n = 76). Diagnostic value was assessed using area under the receiver operating characteristic curve (AUROC), inter-reader reliability using Cohen's κ coefficient.

Results: The consensus optimal cutoff was for %drop ≤ 19.8%, yielding 87.2% accuracy, 95.3% sensitivity, and 73.8% specificity, and that for %fat was ≤ 18.3%, achieving 86.6% accuracy, 96.3% sensitivity, and 70.8% specificity. Both metrics showed high diagnostic performance (AUROC 0.824-0.863) and excellent inter-reader reliability (κ > 0.93, p < 0.001). Multivariate analysis identified %drop ≤ 19.8% and %fat ≤ 18.3% as the strongest independent predictors of malignancy, with odds ratio (OR) being 9.38 and 8.85, respectively (p < 0.001). Signal characteristics on Dixon sequences provided additional diagnostic value, with signal voids on fat-only images (OR 7.14) and high signals on water-only images (OR 5.46).

Conclusion: Quantitative MRI Dixon imaging parameters demonstrated high diagnostic accuracy and excellent inter-reader reliability in differentiating benign and malignant BM lesions, supporting their implementation in clinical practice protocols as a reproducible adjunct to conventional MRI.

Relevance statement: Quantitative Dixon MRI provides reproducible, noninvasive differentiation of bone marrow lesions with high diagnostic accuracy across anatomical sites, enhancing clinical decision-making with standardized thresholds while demonstrating excellent inter-center consistency.

Key points: Quantitative Dixon MRI thresholds of %drop ≤ 19.8% and %fat ≤ 18.3% were established as reliable predictors of malignancy in bone marrow lesions. Dixon metrics demonstrated superior diagnostic accuracy (86.6-87.2%), compared to conventional T1-weighted sequences (79.2%). Excellent inter-reader reliability (κ = 0.895-0.943) supports the reproducibility of quantitative Dixon MRI in clinical practice.

Computed tomography (CT) and magnetic resonance imaging (MRI) are commonly used to assess femoral and tibial torsion. While CT offers high spatial resolution, it involves ionizing radiation. MRI avoids radiation but requires multiple sequences and extended acquisition time. We retrospectively evaluated whether a three-dimensional isotropic MRI localizer (FastView) could serve as a reliable and faster alternative. In this retrospective single-center study, 60 lower limbs from 30 patients, aged 27.1 ± 11.5 years (mean ± standard deviation), 19 females and 11 males, were assessed using both FastView and a dedicated MRI protocol. FastView (5 × 5 × 5 mm³ voxels) imaged the entire lower limb in 17.4 s compared to nearly 7 min for the dedicated protocol. Torsion angles were measured independently by two readers. Agreement between methods was evaluated using intraclass correlation coefficients (ICCs), Bland-Altman plots, and Pearson R². No significant differences in torsion values were found (all p > 0.305). Femoral (ICC: 0.91-0.96) and tibial (ICC: 0.91-0.94) torsion showed excellent inter-modality agreement. Inter-reader reliability was also high (ICC: 0.95-0.99). Correlation values confirmed strong agreement (R²: 0.891-0.963). FastView demonstrated accuracy comparable to the dedicated protocol, offering a fast, efficient, and radiation-free option for routine torsion assessment. RELEVANCE STATEMENT: FastView MRI localizer offers a fast and resource-efficient method for assessing lower limb torsion, potentially replacing standard multisequence protocols in routine clinical practice. KEY POINTS: FastView MRI enables lower limb torsion measurements with full-limb coverage in under 20 s. Torsion angles from FastView and dedicated MRI showed no significant differences. Femoral and tibial ICCs between 0.91 and 0.96 confirm excellent inter-protocol agreement. Inter-reader agreement was consistently high across both protocols. FastView may replace multisequence MRI protocols in routine clinical torsion assessment.