European Radiology Experimental最新文献

Starting from Picasso's quote ("Computers are useless. They can only give you answers"), we discuss the introduction of generative artificial intelligence (AI), including generative adversarial networks (GANs) and transformer-based architectures such as large language models (LLMs) in radiology, where their potential in reporting, image synthesis, and analysis is notable. However, the need for improvements, evaluations, and regulations prior to clinical use is also clear. Integration of LLMs into clinical workflow needs cautiousness, to avoid or at least mitigate risks associated with false diagnostic suggestions. We highlight challenges in synthetic image generation, inherent biases in AI models, and privacy concerns, stressing the importance of diverse training datasets and robust data privacy measures. We examine the regulatory landscape, including the 2023 Executive Order on AI in the United States and the 2024 AI Act in the European Union, which set standards for AI applications in healthcare. This manuscript contributes to the field by emphasizing the necessity of maintaining the human element in medical procedures while leveraging generative AI, advocating for a "machines-in-the-loop" approach.

Background: This study aimed to investigate the long-term effects of repetitive mild traumatic brain injury (rmTBI) with varying inter-injury intervals by measuring diffusion tensor metrics, including mean diffusivity (MD), fractional anisotropy (FA), and diffusion magnitude (L) and pure anisotropy (q).

Methods: Eighteen rats were randomly divided into three groups: short-interval rmTBI (n = 6), long-interval rmTBI (n = 6), and sham controls (n = 6). MD, FA, L, and q values were analyzed from longitudinal diffusion tensor imaging at days 50 and 90 after rmTBI. Immunohistochemical staining against neurons, astrocytes, microglia, and myelin was performed. Analysis of variance, Pearson correlation coefficient, and simple linear regression model were used.

Results: At day 50 post-rmTBI, lower cortical FA and q values were shown in the short-interval group (p ≤ 0.038). In contrast, higher FA and q values were shown for the long-interval group (p ≤ 0.039) in the corpus callosum. In the ipsilesional external capsule and internal capsule, no significant changes were found in FA, while lower L and q values were shown in the short-interval group (p ≤ 0.028) at day 90. The q values in the external capsule and internal capsule were negatively correlated with the number of microglial cells and the total number of astroglial cells (p ≤ 0.035).

Conclusion: Tensor scalar measurements, such as L and q values, are sensitive to exacerbated chronic injury induced by rmTBI with shorter inter-injury intervals and reflect long-term astrogliosis induced by the cumulative injury.

Relevance statement: Tensor scalar measurements, including L and q values, are potential DTI metrics for detecting long-term and subtle injury following rmTBI; in particular, q values may be used for quantifying remote white matter (WM) changes following rmTBI.

Key points: The alteration of L and q values was demonstrated after chronic repetitive mild traumatic brain injury. Changing q values were observed in the impact site and remote WM. The lower q values in the remote WM were associated with astrogliosis.

Introduction: Breast arterial calcifications (BAC) are common incidental findings on routine mammograms, which have been suggested as a sex-specific biomarker of cardiovascular disease (CVD) risk. Previous work showed the efficacy of a pretrained convolutional network (CNN), VCG16, for automatic BAC detection. In this study, we further tested the method by a comparative analysis with other ten CNNs.

Material and methods: Four-view standard mammography exams from 1,493 women were included in this retrospective study and labeled as BAC or non-BAC by experts. The comparative study was conducted using eleven pretrained convolutional networks (CNNs) with varying depths from five architectures including Xception, VGG, ResNetV2, MobileNet, and DenseNet, fine-tuned for the binary BAC classification task. Performance evaluation involved area under the receiver operating characteristics curve (AUC-ROC) analysis, F₁-score (harmonic mean of precision and recall), and generalized gradient-weighted class activation mapping (Grad-CAM++) for visual explanations.

Results: The dataset exhibited a BAC prevalence of 194/1,493 women (13.0%) and 581/5,972 images (9.7%). Among the retrained models, VGG, MobileNet, and DenseNet demonstrated the most promising results, achieving AUC-ROCs > 0.70 in both training and independent testing subsets. In terms of testing F₁-score, VGG16 ranked first, higher than MobileNet (0.51) and VGG19 (0.46). Qualitative analysis showed that the Grad-CAM++ heatmaps generated by VGG16 consistently outperformed those produced by others, offering a finer-grained and discriminative localization of calcified regions within images.

Conclusion: Deep transfer learning showed promise in automated BAC detection on mammograms, where relatively shallow networks demonstrated superior performances requiring shorter training times and reduced resources.

Relevance statement: Deep transfer learning is a promising approach to enhance reporting BAC on mammograms and facilitate developing efficient tools for cardiovascular risk stratification in women, leveraging large-scale mammographic screening programs.

Key points: • We tested different pretrained convolutional networks (CNNs) for BAC detection on mammograms. • VGG and MobileNet demonstrated promising performances, outperforming their deeper, more complex counterparts. • Visual explanations using Grad-CAM++ highlighted VGG16's superior performance in localizing BAC.

Background: Cardiac magnetic resonance (CMR) in the four-chamber plane offers comprehensive insight into the volumetrics of the heart. We aimed to develop an artificial intelligence (AI) model of time-resolved segmentation using the four-chamber cine.

Methods: A fully automated deep learning algorithm was trained using retrospective multicentre and multivendor data of 814 subjects. Validation, reproducibility, and mortality prediction were evaluated on an independent cohort of 101 subjects.

Results: The mean age of the validation cohort was 54 years, and 66 (65%) were males. Left and right heart parameters demonstrated strong correlations between automated and manual analysis, with a ρ of 0.91-0.98 and 0.89-0.98, respectively, with minimal bias. All AI four-chamber volumetrics in repeatability analysis demonstrated high correlation (ρ = 0.99-1.00) and no bias. Automated four-chamber analysis underestimated both left ventricular (LV) and right ventricular (RV) volumes compared to ground-truth short-axis cine analysis. Two correction factors for LV and RV four-chamber analysis were proposed based on systematic bias. After applying the correction factors, a strong correlation and minimal bias for LV volumetrics were observed. During a mean follow-up period of 6.75 years, 16 patients died. On stepwise multivariable analysis, left atrial ejection fraction demonstrated an independent association with death in both manual (hazard ratio (HR) = 0.96, p = 0.003) and AI analyses (HR = 0.96, p < 0.001).

Conclusion: Fully automated four-chamber CMR is feasible, reproducible, and has the same real-world prognostic value as manual analysis. LV volumes by four-chamber segmentation were comparable to short-axis volumetric assessment.

Trials registration: ClinicalTrials.gov: NCT05114785.

Relevance statement: Integrating fully automated AI in CMR promises to revolutionise clinical cardiac assessment, offering efficient, accurate, and prognostically valuable insights for improved patient care and outcomes.

Key points: • Four-chamber cine sequences remain one of the most informative acquisitions in CMR examination. • This deep learning-based, time-resolved, fully automated four-chamber volumetric, functional, and deformation analysis solution. • LV and RV were underestimated by four-chamber analysis compared to ground truth short-axis segmentation. • Correction bias for both LV and RV volumes by four-chamber segmentation, minimises the systematic bias.

Background: Clinical imaging tools to probe aggressiveness of renal masses are lacking, and T2-weighted imaging as an integral part of magnetic resonance imaging protocol only provides qualitative information. We developed high-resolution and accelerated T2 mapping methods based on echo merging and using k-t undersampling and reduced flip angles (TEMPURA) and tested their potential to quantify differences between renal tumour subtypes and grades.

Methods: Twenty-four patients with treatment-naïve renal tumours were imaged: seven renal oncocytomas (RO); one eosinophilic/oncocytic renal cell carcinoma; two chromophobe RCCs (chRCC); three papillary RCCs (pRCC); and twelve clear cell RCCs (ccRCC). Median, kurtosis, and skewness of T2 were quantified in tumours and in the normal-adjacent kidney cortex and were compared across renal tumour subtypes and between ccRCC grades.

Results: High-resolution TEMPURA depicted the tumour structure at improved resolution compared to conventional T2-weighted imaging. The lowest median T2 values were present in pRCC (high-resolution, 51 ms; accelerated, 45 ms), which was significantly lower than RO (high-resolution; accelerated, p = 0.012) and ccRCC (high-resolution, p = 0.019; accelerated, p = 0.008). ROs showed the lowest kurtosis (high-resolution, 3.4; accelerated, 4.0), suggestive of low intratumoural heterogeneity. Lower T2 values were observed in higher compared to lower grade ccRCCs (grades 2, 3 and 4 on high-resolution, 209 ms, 151 ms, and 106 ms; on accelerated, 172 ms, 160 ms, and 102 ms, respectively), with accelerated TEMPURA showing statistical significance in comparison (p = 0.037).

Conclusions: Both high-resolution and accelerated TEMPURA showed marked potential to quantify differences across renal tumour subtypes and between ccRCC grades.

Trial registration: ClinicalTrials.gov, NCT03741426 . Registered on 13 November 2018.

Relevance statement: The newly developed T₂ mapping methods have improved resolution, shorter acquisition times, and promising quantifiable readouts to characterise incidental renal masses.

Sample size, namely the number of subjects that should be included in a study to reach the desired endpoint and statistical power, is a fundamental concept of scientific research. Indeed, sample size must be planned a priori, and tailored to the main endpoint of the study, to avoid including too many subjects, thus possibly exposing them to additional risks while also wasting time and resources, or too few subjects, failing to reach the desired purpose. We offer a simple, go-to review of methods for sample size calculation for studies concerning data reliability (repeatability/reproducibility) and diagnostic performance. For studies concerning data reliability, we considered Cohen's κ or intraclass correlation coefficient (ICC) for hypothesis testing, estimation of Cohen's κ or ICC, and Bland-Altman analyses. With regards to diagnostic performance, we considered accuracy or sensitivity/specificity versus reference standards, the comparison of diagnostic performances, and the comparisons of areas under the receiver operating characteristics curve. Finally, we considered the special cases of dropouts or retrospective case exclusions, multiple endpoints, lack of prior data estimates, and the selection of unusual thresholds for α and β errors. For the most frequent cases, we provide example of software freely available on the Internet.Relevance statement Sample size calculation is a fundamental factor influencing the quality of studies on repeatability/reproducibility and diagnostic performance in radiology.Key points• Sample size is a concept related to precision and statistical power.• It has ethical implications, especially when patients are exposed to risks.• Sample size should always be calculated before starting a study.• This review offers simple, go-to methods for sample size calculations.

Background: Computed tomography (CT) is the usual modality for diagnosing stroke, but conventional CT angiography reconstructions have limitations.

Methods: A phantom with tubes of known diameters and wall thickness was scanned for wall detectability, wall thickness, and contrast-to-noise ratio (CNR) on conventional and spectral black-blood (SBB) images. The clinical study included 34 stroke patients. Diagnostic certainty and conspicuity of normal/abnormal intracranial vessels using SBB were compared to conventional. Sensitivity/specificity/accuracy of SBB and conventional were compared for plaque detectability. CNR of the wall/lumen and quantitative comparison of remodeling index, plaque burden, and eccentricity were obtained for SBB imaging and high-resolution magnetic resonance imaging (hrMRI).

Results: The phantom study showed improved detectability of tube walls using SBB (108/108, 100% versus conventional 81/108, 75%, p < 0.001). CNRs were 75.9 ± 62.6 (mean ± standard deviation) for wall/lumen and 22.0 ± 17.1 for wall/water using SBB and 26.4 ± 15.3 and 101.6 ± 62.5 using conventional. Clinical study demonstrated (i) improved certainty and conspicuity of the vessels using SBB versus conventional (certainty, median score 3 versus 0; conspicuity, median score 3 versus 1 (p < 0.001)), (ii) improved sensitivity/specificity/accuracy of plaque (≥ 1.0 mm) detectability (0.944/0.981/0.962 versus 0.239/0.743/0.495) (p < 0.001), (iii) higher wall/lumen CNR of SBB of (78.3 ± 50.4/79.3 ± 96.7) versus hrMRI (18.9 ± 8.4/24.1 ± 14.1) (p < 0.001), and (iv) excellent reproducibility of remodeling index, plaque burden, and eccentricity using SBB versus hrMRI (intraclass correlation coefficient 0.85-0.94).

Conclusions: SBB can enhance the detectability of intracranial plaques with an accuracy similar to that of hrMRI.

Relevance statement: This new spectral black-blood technique for the detection and characterization of intracranial vessel atherosclerotic disease could be a time-saving and cost-effective diagnostic step for clinical stroke patients. It may also facilitate prevention strategies for atherosclerosis.

Key points: • Blooming artifacts can blur vessel wall morphology on conventional CT angiography. • Spectral black-blood (SBB) images are generated from material decomposition from spectral CT. • SBB images reduce blooming artifacts and noise and accurately detect small plaques.

Presurgical evaluation with functional magnetic resonance imaging (fMRI) can reduce postsurgical morbidity. Here, we discuss presurgical fMRI mapping at ultra-high magnetic fields (UHF), i.e., ≥ 7 T, in the light of the current growing interest in artificial intelligence (AI) and robot-assisted neurosurgery. The potential of submillimetre fMRI mapping can help better appreciate uncertainty on resection margins, though geometric distortions at UHF might lessen the accuracy of fMRI maps. A useful trade-off for UHF fMRI is to collect data with 1-mm isotropic resolution to ensure high sensitivity and subsequently a low risk of false negatives. Scanning at UHF might yield a revival interest in slow event-related fMRI, thereby offering a richer depiction of the dynamics of fMRI responses. The potential applications of AI concern denoising and artefact removal, generation of super-resolution fMRI maps, and accurate fusion or coregistration between anatomical and fMRI maps. The latter can benefit from the use of T1-weighted echo-planar imaging for better visualization of brain activations. Such AI-augmented fMRI maps would provide high-quality input data to robotic surgery systems, thereby improving the accuracy and reliability of robot-assisted neurosurgery. Ultimately, the advancement in fMRI at UHF would promote clinically useful synergies between fMRI, AI, and robotic neurosurgery.Relevance statement This review highlights the potential synergies between fMRI at UHF, AI, and robotic neurosurgery in improving the accuracy and reliability of fMRI-based presurgical mapping.Key points• Presurgical fMRI mapping at UHF improves spatial resolution and sensitivity.• Slow event-related designs offer a richer depiction of fMRI responses dynamics.• AI can support denoising, artefact removal, and generation of super-resolution fMRI maps.• AI-augmented fMRI maps can provide high-quality input data to robotic surgery systems.

Background: We compared magnetic resonance imaging (MRI) turbo spin-echo images reconstructed using a deep learning technique (TSE-DL) with standard turbo spin-echo (TSE-SD) images of the lumbar spine regarding image quality and detection performance of common degenerative pathologies.

Methods: This prospective, single-center study included 31 patients (15 males and 16 females; aged 51 ± 16 years (mean ± standard deviation)) who underwent lumbar spine exams with both TSE-SD and TSE-DL acquisitions for degenerative spine diseases. Images were analyzed by two radiologists and assessed for qualitative image quality using a 4-point Likert scale, quantitative signal-to-noise ratio (SNR) of anatomic landmarks, and detection of common pathologies. Paired-sample t, Wilcoxon, and McNemar tests, unweighted/linearly weighted Cohen κ statistics, and intraclass correlation coefficients were used.

Results: Scan time for TSE-DL and TSE-SD protocols was 2:55 and 5:17 min:s, respectively. The overall image quality was either significantly higher for TSE-DL or not significantly different between TSE-SD and TSE-DL. TSE-DL demonstrated higher SNR and subject noise scores than TSE-SD. For pathology detection, the interreader agreement was substantial to almost perfect for TSE-DL, with κ values ranging from 0.61 to 1.00; the interprotocol agreement was almost perfect for both readers, with κ values ranging from 0.84 to 1.00. There was no significant difference in the diagnostic confidence or detection rate of common pathologies between the two sequences (p ≥ 0.081).

Conclusions: TSE-DL allowed for a 45% reduction in scan time over TSE-SD in lumbar spine MRI without compromising the overall image quality and showed comparable detection performance of common pathologies in the evaluation of degenerative lumbar spine changes.

Relevance statement: Deep learning-reconstructed lumbar spine MRI protocol enabled a 45% reduction in scan time compared with conventional reconstruction, with comparable image quality and detection performance of common degenerative pathologies.

Key points: • Lumbar spine MRI with deep learning reconstruction has broad application prospects. • Deep learning reconstruction of lumbar spine MRI saved 45% scan time without compromising overall image quality. • When compared with standard sequences, deep learning reconstruction showed similar detection performance of common degenerative lumbar spine pathologies.

Background: The potential role of cardiac computed tomography (CT) has increasingly been demonstrated for the assessment of diffuse myocardial fibrosis through the quantification of extracellular volume (ECV). Photon-counting detector (PCD)-CT technology may deliver more accurate ECV quantification compared to energy-integrating detector CT. We evaluated the impact of reconstruction settings on the accuracy of ECV quantification using PCD-CT, with magnetic resonance imaging (MRI)-based ECV as reference.

Methods: In this post hoc analysis, 27 patients (aged 53.1 ± 17.2 years (mean ± standard deviation); 14 women) underwent same-day cardiac PCD-CT and MRI. Late iodine CT scans were reconstructed with different quantum iterative reconstruction levels (QIR 1-4), slice thicknesses (0.4-8 mm), and virtual monoenergetic imaging levels (VMI, 40-90 keV); ECV was quantified for each reconstruction setting. Repeated measures ANOVA and t-test for pairwise comparisons, Bland-Altman plots, and Lin's concordance correlation coefficient (CCC) were used.

Results: ECV values did not differ significantly among QIR levels (p = 1.000). A significant difference was observed throughout different slice thicknesses, with 0.4 mm yielding the highest agreement with MRI-based ECV (CCC = 0.944); 45-keV VMI reconstructions showed the lowest mean bias (0.6, 95% confidence interval 0.1-1.4) compared to MRI. Using the most optimal reconstruction settings (QIR4. slice thickness 0.4 mm, VMI 45 keV), a 63% reduction in mean bias and a 6% increase in concordance with MRI-based ECV were achieved compared to standard settings (QIR3, slice thickness 1.5 mm; VMI 65 keV).

Conclusions: The selection of appropriate reconstruction parameters improved the agreement between PCD-CT and MRI-based ECV.

Relevance statement: Tailoring PCD-CT reconstruction parameters optimizes ECV quantification compared to MRI, potentially improving its clinical utility.

Key points: • CT is increasingly promising for myocardial tissue characterization, assessing focal and diffuse fibrosis via late iodine enhancement and ECV quantification, respectively. • PCD-CT offers superior performance over conventional CT, potentially improving ECV quantification and its agreement with MRI-based ECV. • Tailoring PCD-CT reconstruction parameters optimizes ECV quantification compared to MRI, potentially improving its clinical utility.