Frontiers in radiology最新文献

Introduction: Adrenocortical carcinoma presents significant diagnostic challenges due to its histological heterogeneity and variable clinical behavior. This study aimed to evaluate the diagnostic value of radiomic features in predicting mitotic activity (low/high-grade) and morphological variants (conventional, oncocytic, myxoid) of adrenocortical carcinoma.

Materials and methods: A retrospective analysis of 32 patients with histologically confirmed ACC (18 conventional, 9 oncocytic and 5 myxoid cases) was performed, with mitotic data available for 25 cases (13 low-grade and 12 high-grade cases). Radiomic features including Gray-Level Co-occurrence Matrix (GLCM), Run-Length (GLRLM), Size-Zone (GLSZM), Dependence (GLDM), Neighboring-Tone (NGTDM) and first order features were extracted from four-phase CT using PyRadiomics after manual 3D segmentation. Statistical analysis included Mann-Whitney U, Kruskal-Wallis tests, ROC curve (AUC, sensitivity, specificity) and PPV, NPV assessment.

Results: Our analysis demonstrated statistically significant differences between tumor grades with firstorder_Skewness (AUC = 0.924, 95% CI: 0.819-0.986; p = 0.005) showing high predictive performance in the venous phase. Radiomic features did not show statistically significant differences between morphological variants of ACC after adjustment for multiple comparisons.

Conclusion: Our results confirm the value of CT radiomics for preoperative stratification of ACC grade, but the question of differentiation of morphological variants remains unresolved and requires further validation in larger cohorts.

Introduction: Neurofibromatosis type 2 related Schwannomatosis (NF2-SWN) is a genetic disorder characterized by the growth of vestibular schwannomas (VS), which often leads to progressive hearing loss and vestibular dysfunction. Accurate volumetric assessment of VS tumors is crucial for effective monitoring and treatment planning. Since tumor growth dynamics are often subtle, the resolution of MRI scans plays a critical role in detecting small volumetric changes that inform clinical decisions. This study evaluates the impact of MRI voxel resolution on the accuracy of manual and AI-driven volumetric segmentation of VS in NF2-SWN patients.

Methods: Ten patients with NF2-SWN, totaling 17 tumors, underwent high-resolution MRI scans with varying voxel sizes on different MRI machines at Yale New Haven Hospital. Tumors were segmented using both manual and AI-based methods, and the effect of voxel size on segmentation precision was quantified through volume measurements, Dice similarity coefficients, and Hausdorff distances.

Results: Results indicate that larger voxel sizes (1.2 × 0.9 × 4.0 mm) significantly reduced segmentation accuracy when compared to smaller voxel sizes (0.5 × 0.5 × 0.8 mm). In addition, AI-based segmentation outperformed manual methods, particularly at larger voxel sizes.

Discussion: These findings highlight the importance of optimizing voxel resolution for accurate tumor monitoring and suggest that AI-driven segmentation may improve consistency and precision in NF2-SWN tumor surveillance.

Future AI systems may need to provide medical professionals with explanations of AI predictions and decisions. While current XAI methods match these requirements in principle, they are too inflexible and not sufficiently geared toward clinicians' needs to fulfill this role. This paper offers a conceptual roadmap for how XAI may be integrated into future medical practice. We identify three desiderata of increasing difficulty: First, explanations need to be provided in a context- and user-dependent manner. Second, explanations need to be created through a genuine dialogue between AI and human users. Third, AI systems need genuine social capabilities. We use an imaginary stroke treatment scenario as a foundation for our roadmap to explore how the three challenges emerge at different stages of clinical practice. We provide definitions of key concepts such as genuine dialogue and social capability, we discuss why these capabilities are desirable, and we identify major roadblocks. Our goal is to help practitioners and researchers in developing future XAI that is capable of operating as a participant in complex medical environments. We employ an interdisciplinary methodology that integrates medical XAI, medical practice, and philosophy.

Background: The aim of this systematic review was to determine the patency and complications related to percutaneous metallic biliary stent placement for malignant biliary obstruction in the current literature.

Methods: This review was performed using the Preferred Reporting Items of Systematic Reviews and Meta-Analyses guidelines. EMBASE and PubMed were queried yielding 891 articles, 18 of which were included in the final analysis. The Newcastle-Ottawa Quality Assessment Scale was used to appraise article quality. Patient demographics, technical success rate, and procedure outcomes were recorded. Complications were classified as "major" if they resulted in blood transfusion or additional invasive procedures or were reported as such in the literature. Complications that did not meet these criteria were classified as "minor".

Results: A total of 1,453 patients (677 female; weighted age 66.8 years) underwent biliary stent placement. The weighted technical success rate was 97.7%. The incidence of stent occlusion was 13.5% with 6.6% of patients requiring further intervention to maintain patency. There were 277 (19.1%) complications, of which 87 were classified as major. The most common complications were pancreatitis (93, 6.4%), cholangitis (69, 4.8%), and bleeding (64, 4.4%). In cases of bleeding, 4.7% of patients needed a blood transfusion and 15.6% required a procedure to treat bleeding. There were 6 (0.4%) procedure-related deaths.

Conclusion: In conclusion, percutaneous metallic stent placement for malignant biliary obstruction has a high technical success rate and relatively low rate of occlusion. Although nearly one in five procedures resulted in a complication, most cases were minor.

Objective: This study constructs a deep learning-based combined algorithm named WaveAttention ResNet (WARN) to investigate the classification accuracy for seven common retinal diseases and the feasibility of AI-assisted diagnosis in this field.

Methods: First, a deep learning-based classification network is constructed. The network is built upon ResNet18, integrated with the Convolutional Block Attention Module (CBAM) and wavelet convolution modules, forming the WARN method for retinal disease classification. Second, the public OCTDL dataset is used to train WARN, which contains classification data for seven retinal disease types: age-related macular degeneration (AMD), diabetic macular edema (DME), epiretinal membrane (ERM), normal (NO), retinal artery occlusion (RAO), retinal vein occlusion (RVO), and vitreomacular interface disease (VID). During this process, ablation experiments and significance tests are conducted on WARN, and comprehensive analyses of various indicators for WARN, ResNet-18, ResNet-50, Swin Transformer v2, EfficientNet, and Vision Transformer (ViT) are performed in retinal disease classification tasks. Finally, data provided by Shanxi Eye Hospital are used for testing, and classification results are analyzed.

Results: WARN demonstrates excellent performance on the public OCTDL dataset. Ablation experiments and significance tests confirm the effectiveness of WARN, achieving an accuracy of 90.68%, F1-score of 91.29%, AUC of 97.50%, precision of 93.31%, and recall of 90.68% with relatively short training time. In the dataset from Shanxi Eye Hospital, WARN also performs well, with a recall of 90.85%, precision of 79.94%, and accuracy of 89.18%.

Conclusion: This study fully confirms that the constructed WARN is efficient and feasible for classifying seven common retinal diseases. It further highlights the enormous potential and broad application prospects of AI technology in the field of auxiliary medical diagnosis.

Transarterial microembolization (TAME) has gained recognition as a minimally invasive treatment for chronic musculoskeletal pain, demonstrating significant efficacy with a favorable safety profile ( 1, 2). However, complications remain underreported. This case report describes the first documented severe adverse event in a patient with a chronic venous ulcer undergoing TAME for a micro arteriovenous fistula (AVF). The patient developed significant complications, including extensive leg swelling, skin changes, and cellulitis requiring prolonged inpatient care. These findings highlight the importance of patient selection and embolic agent considerations to mitigate potential risks associated with TAME.

Intoduction: The palatal rugae have been suggested to be just as unique as the human fingerprint. Therefore, endeavors have been made to utilize this uniqueness for the identification of disaster victims. With the rise of digital 3D dental data, computational comparisons of palatal rugae have become possible. But a direct comparison of the full palatal scan by iterative closest point (ICP) has shown to be tedious and demands a knowledge of superimposition software.

Methods: Here, we propose (1) an automatic extraction of the palatal rugae ridges from the 3D scans, followed by (2) ICP of the extracted ridges.

Results: Pairwise comparisons of palates take less than a second, and in this study, it was possible to distinguish between palates from the same individual vs. palates from different individuals with a receiver operating characteristic area-under-the-curve of 0.994.

Discussion: This shows that the extraction of the palatal rugae ridges is a potential efficient addition to the toolbox of a forensic odontologist for disaster victim identification.

Background: Previous studies have reported that quantum noise inherently present in CT images hinders the generation of CT-based ventilation image (CTVI), while quantum noise reduction approaches that do not affect CTVI have not yet been reported.

Aims: The purpose of this study was to evaluate the impact of noise reduction preprocessing on the accuracy and robustness of CTVI in relation to quantum noise present in CT images.

Methods and material: To reproduce the quantum noise, Gaussian noise (SD: 30, 80, 150 HU) was added to each inhalation and exhalation CT image. CTVI_ref and CTVI_noise was generated from CT_ref and CT_noise. A median filter and the noise reduction by the CNN were also applied to the CT image, which contained the quantum noise, and CTVI_med and CTVI_cnn was created in the same manner as CTVI_ref. We evaluated whether the regions classified as high, middle, or low in CTVI_ref were accurately represented as high, middle, or low in CTVI_noise, CTVI_med and CTVI_cnn. Additionally, to evaluate the ventilation function of each voxel, we compared two-dimensional histograms of CTVI_ref, CTVI_noise, CTVI_med and CTVI_cnn.

Statistical analysis used: Cohen's kappa coefficient and Spearman's correlation were used to assess the agreement between CTVI_ref and each of the following: CTVI_noise, CTVI_med, and CTVI_cnn.

Results: CTVI_cnn significantly improved categorical consistency and voxel-level correlation of CTVI, particularly under high-noise conditions (150 HU), outperforming both CTVI_noise and CTVI_med.

Conclusions: CNN-based denoising effectively improved the accuracy and robustness of CTVI under quantum noise.