Insights into Imaging最新文献

Diagnostic errors in mammography-particularly missed or delayed breast cancer detection-have a substantial impact on patient outcomes. These misdiagnoses remain a leading cause of malpractice claims in radiology, underscoring their serious clinical and legal implications. Contributing factors to errors in breast imaging include reader-related cognitive biases, lesion characteristics, patient-specific variables, and technical limitations. To address these challenges, a systematic approach is essential. Key strategies include structured error recognition, peer review processes, and robust quality assurance programs. Educational initiatives and system-level interventions-such as structured training, continuous feedback loops, and the integration of AI-driven computer-aided detection (CAD) tools-can significantly reduce diagnostic errors and enhance accuracy in breast imaging interpretation. This article aims to highlight common pitfalls in mammography, analyze root causes, and propose practical strategies for improvement. Real-life cases of missed diagnoses are included to reinforce key learning points and support radiologists in improving diagnostic precision and improving patient care. CRITICAL RELEVANCE STATEMENT: Missed or delayed breast cancer diagnoses stem from multiple factors. A multi-pronged strategy-combining peer review, bias mitigation, education, supportive environments, and AI tools-can improve diagnostic accuracy and enhance interpretive accuracy and advance quality standards in breast imaging practice. KEY POINTS: Missed or delayed breast cancer diagnoses on mammography continue to be a significant source of diagnostic error with serious clinical and medico-legal consequences. Contributing factors to missed or delayed breast cancer diagnoses include cognitive biases, subtle lesion characteristics, patient-specific variables, and technical limitations. Structured peer review, double reading, and robust quality assurance programs can reduce interpretive variability and improve diagnostic performance. Educational initiatives and AI-driven tools, such as computer-aided detection (CAD), support error reduction and enhance accuracy in breast imaging interpretation.

Objectives: To establish a metabolic burden-based clinical-radiological model for predicting postoperative recurrence in hepatitis B virus (HBV)-related hepatocellular carcinoma (HCC) at Barcelona Clinic Liver Cancer (BCLC) stages 0-A.

Materials and methods: This retrospective multi-center study included HBV-related HCC (BCLC 0-A) undergoing curative surgery. Metabolic burden was defined as the cumulative number of metabolic abnormalities. Trend test assessed dose-dependent relationship. Predictors were identified via univariate and multivariate Cox regression analyses, and a nomogram was developed. The model underwent internal validation (5-fold, 100 times cross) and external validation. Performance was evaluated using C-index, calibration curves, and decision curve analysis.

Results: The internal and external cohorts consisted of 363 patients (55.9 ± 10.7 years, 295 males) and 74 patients (55.5 ± 10.2 years, 55 males). Recurrence risk increased by 1.53 times (p = 0.049) and 1.64 times (p = 0.018) for patients with 2 and 3-4 metabolic abnormalities (ptrend = 0.022). Independent predictors included tumor burden score > 2.4 (HR = 2.40, p = 0.003), metabolic abnormalities ≥ 2 (HR = 1.49, p = 0.023), aspartate transaminase/alanine transaminase ratio > 1 (HR = 1.51, p = 0.012), albumin-bilirubin grade 2 (HR = 1.70, p = 0.020), arterial rim enhancement (HR = 1.87, p = 0.002) and mosaic appearance (HR = 1.55, p = 0.033). C-indices for predicting 2- and 5-year recurrence were 0.728 (95% CI: 0.726-0.729) and 0.674 (95% CI: 0.673-0.675) in training sets, 0.716 (95% CI: 0.711-0.720) and 0.657 (95% CI: 0.653-0.660) in internal validation sets, and 0.710 (95% CI: 0.602-0.855) and 0.683 (95% CI: 0.594-0.798) in external cohort. The model showed higher predictive efficacy (p < 0.001 for all) and better clinical net benefit compared to BCLC and CNLC staging systems in the very early/early-stage of HCCs.

Conclusion: The metabolic burden-based clinical-radiological model effectively predicts postoperative recurrence in HBV-related HCC.

Critical relevance statement: Patients with HBV-related HCC who have two or more coexisting metabolic abnormalities may have a higher risk of postoperative recurrence. The metabolic burden-based clinical-radiological model is valuable in predicting postoperative recurrence KEY POINTS: Metabolic abnormalities were dose-dependently related to the risk of postoperative recurrence. The clinical-radiological model showed well-predictive efficacy in validation cohorts. The clinical-radiological model displayed higher efficacy compared to existing staging systems for the very early/early-stage of HCCs.

Objectives: This study aimed to evaluate the efficacy of two combined ultrafast breast MRI kinetic parameters, combination 1 including time to enhancement [TTE], maximum slope [MS], and initial enhancement phase [IE phase] compared to combination 2 including relative enhancement [RE], maximum enhancement [ME], maximum relative enhancement [MRE], time to peak [TTP], and wash in rate in characterizing benign and malignant breast lesions.

Materials and methods: This prospective study included 264 female patients with 273 breast lesions. The ultrafast protocol was done using the TWIST sequence. The parameters for combination 1 were generated manually; however, the parameters for combination 2 were generated semi-automatically. The overall performance of the ultrafast protocol was compared to the conventional dynamic MRI protocol.

Results: The ultrafast protocol was obtained in 77 s. The mean interpretation time was 5 ± 2.7 and 1 ± 0.5 min for combinations 1 and 2, respectively. Combination 1 showed an AUC of 0.910, a sensitivity of 76.5% and a specificity of 90%, while combination 2 showed an AUC of 0.869, a sensitivity of 76.5%, and a specificity of 85% in differentiating benign from malignant lesions. Upon combining all parameters, the AUC, sensitivity, and specificity in discriminating between the two groups increased to 0.944, 80.4%, and 85%, respectively. Both ultrafast techniques and conventional MRI demonstrated excellent performance in discriminating between benign and malignant lesions (AUC = 0.921 vs 0.940, respectively).

Conclusion: Adding the semiautomatically generated parameters derived from ultrafast breast MRI can improve the performance in characterizing breast lesions.

Critical relevance statement: By studying ultrafast-derived semiautomatic, easily applicable parameters, we aim to reduce the acquisition and interpretation times of breast MRI without compromising performance, when used as a problem-solving modality in indeterminate breast lesions to characterize them as either benign or malignant.

Key points: Adding semiautomatic ultrafast parameters to the MS and TTE improves the overall performance in characterizing breast lesions. The combined ultrafast parameters provide the highest discriminating power between benign and malignant breast lesions. Ultrafast MRI showed comparable performance to conventional dynamic contrast-enhanced MRI in the discrimination between benign and malignant breast lesions.

Objectives: To develop multivariate models for preoperative detection of occult lymph node (LN) metastasis in gastric cancer (GC), by integrating CT-based extracellular volume (ECV) fraction and clinicopathological features, and further evaluate the prognostic value of the combined model.

Materials and methods: This retrospective study included 129 GCs with the N (-) group (n = 49) and the N (+) group (n = 80). The preoperative CT parameters (including ECV fraction), WHO types and differentiation degree based on endoscopic pathological, and 4 hematological indices were assessed. The diagnostic performance of multivariate models was evaluated by receiver operating characteristic curve analysis.

Results: The N (+) group demonstrated significantly higher proportions of poorly cohesive carcinoma and poor differentiation based on endoscope (both p < 0.001). Significantly higher CT-measured tumor area and ECV fraction were seen in the N (+) group (p < 0.001 and p = 0.008, respectively), and a significantly higher proportion of ECV value > 50% in the N (+) group (p = 0.001). The clinicopathological model, CT parameters model, and combined model yielded areas under the curves of 0.768, 0.774, and 0.843, respectively. The combined model with the high-risk group revealed a significantly shorter median recurrence-free survival compared to the low-risk group (p = 0.008).

Conclusion: The proposed preoperative combined model exhibited a promising performance for early predicting occult LN metastasis and stratifying postoperative recurrence risk in GC, by integrating CT-based ECV fraction and clinicopathological features.

Critical relevance statement: The CT-based ECV preoperative model could potentially provide valuable clinical reference for making clinical strategies in GC.

Key points: It is a great challenge for clinicians to evaluate occult lymph node (LN) status in gastric cancer (GC). The N (+) group demonstrated higher CT-based extracellular volume (ECV) fractions and tumor area, and higher proportions of poorly cohesive carcinoma and poor differentiation. This model helped preoperative detection of occult LN metastasis and stratifying postoperative recurrence risk in GC.

Objective: The study sought to develop and validate an MRI-based deep learning radiomics (DLR) nomogram for preoperative prediction of vessels encapsulating tumor clusters (VETC) and recurrence-free survival (RFS) in hepatocellular carcinoma (HCC).

Materials and methods: The dual-center study retrospectively enrolled 625 HCC patients who underwent preoperative Gd-EOB-DTPA-enhanced MRI, including training (n = 296), internal (n = 126), and external (n = 203) test sets. Clinical-radiologic characteristics were selected to develop a clinical-radiologic model. Habitat radiomics and deep learning (DL) features were extracted and selected to develop the habitat radiomics and DL models using the machine learning classifiers. The DLR nomogram model was ultimately constructed by integrating univariate-selected clinical-radiologic characteristics with habitat radiomics and DL scores. Both univariable and multivariable Cox regression analyses were performed to identify independent prognostic factors and develop a prognostic model for RFS.

Results: In the external test set, the DLR nomogram model yielded a higher area under the curve (AUC) than the clinical-radiologic model (0.752 vs 0.678; p = 0.004), while habitat radiomics (0.750) and DL models (0.748) showed comparable performance (both p > 0.05). The DLR nomogram consistently demonstrated the higher F1-scores across all three sets. The prognostic model incorporating AFP (hazard ratio (HR), 1.628 [95% CI: 1.113-2.380]; p = 0.012) and DLR score (1.279 [1.051-1.557]; p = 0.014) achieved C-indexes of 0.679 and 0.642 for RFS in the internal and external test sets.

Conclusion: The DLR nomogram model helps predict VETC in HCC and assess the risk for RFS.

Critical relevance statement: Interpretable deep learning radiomics nomogram model provides clinicians with more precise technical support for preoperative prediction of VETC status and RFS in HCC, potentially aiding in clinical decision-making and follow-up strategies.

Key points: Vessels encapsulating tumor clusters (VETC) is a critical predictor of aggressive hepatocellular carcinoma. The deep learning radiomics (DLR) nomogram model helps predict VETC, and the DLR score serves as an independent prognostic factor for recurrence-free survival. The model demonstrated favorable interpretability through the SHAP method.

Magnetic resonance imaging (MRI) of the neuroaxis is prone to a variety of artifacts. Familiarity with these artifacts and their respective mitigation techniques is essential for accurate neuroradiological interpretation. In this educational review, we focus on artifacts caused by the physiological flow of cerebrospinal fluid (CSF), which are encountered commonly and, depending on the context, may be beneficial or detrimental in diagnostic decision-making. The pictorial examples provided will illustrate key cases with their practical implications. CRITICAL RELEVANCE STATEMENT: This paper highlights common CSF flow artifacts, including phase encoding artifacts, time-of-flight signal loss, entry slice phenomenon, and intravoxel dephasing, emphasizing their impact on diagnosis interpretation and mitigation strategies. KEY POINTS: CSF artifacts stem from flow dynamics, phase differences, or magnetic field interactions. Artifacts obscure or mimic pathologies, degrade image quality, or occasionally aid in diagnostic decision-making. Mitigation strategies are simple and intuitive, including modification of phase directions, employing alternate imaging sequences, and altering MRI parameters.

Objectives: To investigate normal and pathologic values of fatty infiltration (FI) and muscle volume through volumetric quantification of the main hip abductors of patients with unilateral greater trochanteric pain syndrome (GTPS) using 2-point-Dixon MRI.

Materials and methods: Patients prospectively underwent MRI of both hips: FI of the gluteus minimus (Gmin) and medius (Gmed) muscles were quantified by volumetric fat fractions (3D FF) using 2-point-Dixon MRI. Whole (WMV) and lean muscle volumes (LMV) were calculated for both muscles. 3D FF and volumes were compared between asymptomatic and GTPS hips, using the Wilcoxon signed-rank test. Gender-specific differences were assessed using the Mann-Whitney U test.

Results: Forty-one patients (mean age 65.0 ± 13.6 years, 27 females) were analyzed. 3D FF in asymptomatic hips was lower than in symptomatic hips (Gmin: 17.8% vs. 19.8%; Gmed: 12.7% vs. 15.9% (all p ≤ 0.02)). Gmin had a higher 3D FF than Gmed (p < 0.001). Females had higher FF (asymptomatic and symptomatic Gmin: 19.4%, 21.8%; asymptomatic and symptomatic Gmed: 13.2%, 16.3%) than males (asymptomatic and symptomatic Gmin: 14.7%, 16.1%; asymptomatic and symptomatic Gmed: 11.8%, 14.9%) for both sides and muscles. Average WMV in asymptomatic hips for Gmin and Gmed were 77.2 cm³, 270.1 cm³ in females, and lower in males (both p < 0.001) with 107.1 cm³, 408.1 cm³, respectively.

Conclusion: This study offers reference values for 3D FF and volumes of the Gmin and Gmed muscles in asymptomatic elderly hips, which are significantly lower than their GTPS counterparts, with succinctly higher fat fractions in females than males. Women showed significantly lower muscle volume for both muscles than men.

Critical relevance statement: Volumetric fat fractions of gluteal muscles show significant symptoms and gender related differences, indicating their potential as an imaging biomarker in the common GTPS patient.

Key points: In females, asymptomatic hips showed average volumetric fat fractions of 19% for Gmin and 13% for Gmed; with lower values in males, of 15% and 12%, respectively. Whole muscle volumes in asymptomatic hips for Gmin and Gmed were 77.2 cm³, 270.1 cm³ in females, and 107.1 cm³, 408.1 cm³ in males. Using volumetric fat fractions, abductor muscle fat content was significantly higher in symptomatic GTPS hips compared to asymptomatic hips.

Objectives: High classification accuracy of Alzheimer's disease (AD) from structural MRI has been achieved using deep neural networks, yet the specific image features contributing to these decisions remain unclear. In this study, the contributions of T1-weighted (T1w) gray-white matter texture, volumetric information, and preprocessing-particularly skull-stripping-were systematically assessed.

Materials and methods: A dataset of 990 matched T1w MRIs from AD patients and cognitively normal controls from the ADNI database was used. Preprocessing was varied through skull-stripping and intensity binarization to isolate texture and shape contributions. A 3D convolutional neural network was trained on each configuration, and classification performance was compared using exact McNemar tests with discrete Bonferroni-Holm correction. Feature relevance was analyzed using Layer-wise Relevance Propagation, image similarity metrics, and spectral clustering of relevance maps.

Results: Despite substantial differences in image content, classification accuracy, sensitivity, and specificity remained stable across preprocessing conditions. Models trained on binarized images preserved performance, indicating minimal reliance on gray-white matter texture. Instead, volumetric features-particularly brain contours introduced through skull-stripping-were consistently used by the models.

Conclusion: This behavior reflects a shortcut learning phenomenon, where preprocessing artifacts act as potentially unintended cues. The resulting Clever Hans effect emphasizes the critical importance of interpretability tools to reveal hidden biases and to ensure robust and trustworthy deep learning in medical imaging.

Critical relevance statement: We investigated the mechanisms underlying deep learning-based disease classification using a widely utilized Alzheimer's disease dataset, and our findings reveal a reliance on features induced through skull-stripping, highlighting the need for careful preprocessing to ensure clinically relevant and interpretable models.

Key points: Shortcut learning is induced by skull-stripping applied to T1-weighted MRIs. Explainable deep learning and spectral clustering estimate the bias. Highlights the importance of understanding the dataset, image preprocessing and deep learning model, for interpretation and validation.