Non-contrast MRI (bi-parametric MRI-bpMRI) has been investigated as a potential tool to be integrated in clinically significant prostate cancer (csPCa) screening. Moreover, artificial intelligence (AI) is emerging too as a potential support, especially for less-experienced radiologists. Therefore, the aim of this study was to evaluate the effectiveness of an AI-based software in csPCa screening using bpMRI, with a focus on supporting less-experienced radiologists. A retrospective analysis was conducted within the PROSA-trial, a randomized, single-center study involving 759 men eligible for PCa screening. BpMRI were acquired using prostate imaging reporting and data system (PI-RADS) v2.1-compliant protocols and evaluated independently by an expert radiologist, a less-experienced reader, AI-based software, and the less-experienced reader with AI support. Diagnostic performance was assessed using ROC curves and inter-reader agreement (Cohen's kappa), using expert interpretation as the reference standard. Four hundred ninety-nine bpMRI were analyzed. The AI-assisted less-experienced reader achieved the highest diagnostic performance (sensitivity 76.5%, specificity 97.2%, accuracy 95.8%, AUC 0.969), surpassing both AI-alone (sensitivity 58.8%, specificity 96.6%, accuracy 94.0%, AUC 0.952) and unaided less-experienced reader (sensitivity 67.6%, specificity 95.1%, accuracy 93.2%, AUC 0.932). Inter-reader agreement improved with AI assistance (from κ = 0.64 to 0.84). AI assistance reduced equivocal PI-RADS 3 cases (from 77 to 53) and improved exact agreement with the expert from 32.5% to 54.5%, while also reducing diagnostic discordance. AI can support less experienced radiologists and enhance consistency in bpMRI interpretation, especially considering equivocal cases. Moreover, integrating AI into radiology workflows can alleviate reporting burden and help prioritize suspicious cases, offering critical advantages in high-volume PCa screening settings. Question Can AI improve the diagnostic performance and consistency of less experienced radiologists interpreting non-contrast prostate MRI in a screening setting? Findings AI assistance significantly improved effectiveness and inter-reader agreement in prostate MRI interpretation, particularly for less experienced radiologists within a screening population. Clinical relevance Integrating AI into prostate MRI workflows may enhance screening efficiency, reduce variability, and support equitable early detection of csPCa, with a positive impact on prioritization of reporting.

Involvement of the bone marrow by metastases fromsolid tumors or multiple myeloma (MM) is a critical challenge in oncologic imaging. Lesion detection and staging, as well as accurate assessment of treatment response, disease recurrence, and complications, are key to optimal patient management. This article provides recommendations for performing and interpreting bone marrow MRI in cancer patients. MRI should be the primary imaging modality for patients suspected of having skeletal bone metastases or MM, and should replace radiography, bone scintigraphy, and CT for these indications. Protocols must be tailored to the clinical context and to each specific cancer. Whole-body MRI (WB-MRI) is preferred for a comprehensive assessment, while axial skeleton MRI (AS-MRI) is a fast and reliable alternative for targeted or follow-up evaluations. We recommend standardized protocols that incorporate anatomical sequences (preferably fast spin echo T2 Dixon) and diffusion-weighted imaging (DWI). Quantitative biomarkers, e.g., apparent diffusion coefficient (ADC) and fat fraction (FF), should be implemented to improve diagnostic accuracy and evaluate treatment response. Radiologists must be familiar with the typical patterns of bone marrow replacement by cancer cells, response assessment principles, and common imaging pitfalls. Every medical imaging facility should offer optimal bone marrow MRI and implement these recommendations using available MRI systems and existing disease-oriented guidelines. This ESR Essentials illustrates when, how, and why to perform bone marrow MRI to improve diagnostic precision and oncologic care across a broad range of indications. KEY POINTS: Prefer MRI of the bone marrow over radiographs, bone scintigraphy, or CT for suspected bone metastases of solid cancers and for myeloma staging. Use MRI for diagnosis of bone involvement, disease staging, assessment of lesion response to treatment, detection of recurrence, and assessment of osseous complications. Tailor MRI protocol to cancer type following existing guidelines, targeting either the axial skeleton or the "whole body," and using a panel of sequences with fat-sensitive, fast spin echo T2 Dixon and diffusion-weighted sequences as the fundamental components.

Early recognition of individuals at elevated risk for new ipsilateral ischemic lesions (NIILs) after carotid artery stenting (CAS) is vital for planning effective preventive interventions. The aim of this study was to develop a deep learning (DL) radiomics model to predict NIILs post-CAS from dual-energy CT (DECT) images. This study retrospectively enrolled patients from three centers. Carotid plaques were delineated on multiparametric DECT images. A combined model integrating clinical-radiological, handcrafted radiomics (HCR), and DL features was constructed using a support vector machine algorithm to predict NIILs. The model's performance was assessed through the area under the receiver operating characteristic curve (AUC). To improve the interpretability of the model, SHapley Additive exPlanations (SHAP) analysis was applied. This study involved 336 patients divided into the training (n = 135), internal validation (n = 58), and external test (n = 143) cohorts. NIILs were present in 38.5%, 37.9%, and 39.9% of the subjects, respectively. Symptomatic events and plaque ulceration were identified as independent risk factors for NIILs. The combined model incorporating 2 clinical-radiological risk factors, 9 HCR features, and 15 DL features demonstrated satisfactory performance in predicting NIILs, with AUCs of 0.908, 0.842, and 0.856 in the three cohorts, respectively. The predictions of the combined model were explained both locally and globally by SHAP analysis. The combined model demonstrated high accuracy in identifying patients at elevated risk for NIILs post-CAS and can serve as an interpretable tool for optimizing treatment strategies. Question Early prediction of new ipsilateral ischemic lesions (NIILs) after carotid artery stenting (CAS) is crucial for timely interventions, but no effective, interpretable predictive method exists. Findings The combined model incorporating deep learning radiomics features extracted from multiparametric dual-energy CT images and clinical-radiological features demonstrated high accuracy in predicting NIILs after CAS. Clinical relevance The combined model offers an interpretable tool for identifying patients at high risk for NIILs post-CAS, potentially improving personalized treatment strategies and patient outcomes by enabling targeted preventive care.

Microvascular invasion is a strong prognostic factor in hepatocellular carcinomas. The aim of our study was to assess the diagnostic value of mechanical parameters measured with compression MR elastography to detect microvascular invasion in hepatocellular carcinomas. In this prospective preoperative MR elastographic study, consecutive patients with hepatocellular carcinomas, scheduled for tumor surgical resection, were included. The tumor parameters assessed with MR elastography were the basal visco-elastic parameters (storage modulus, loss modulus, and phase angle, reflecting elasticity, viscosity and visco-elastic ratio) during expiration and inspiration, and the tumor stiffening slope during compression induced by respiration, reflecting non-linear elasticity. Microvascular invasion was determined with histopathological examination of resected tumors. Diagnostic performance of MR elastography was assessed with area under the receiver operating curve (AUC) analysis. The final study group consisted of 53 patients with complete surgical resection, MR elastography and histological data, including 31 patients with microvascular invasion. Compression stiffening slope and storage modulus difference between inspiration and expiration were significantly higher in hepatocellular carcinomas without than with microvascular invasion (p < 0.001 and p = 0.03, respectively). Among clinical, morphological and biomechanical imaging features, the MR elastography compression stiffening slope (p = 0.004) and histological WHO differentiation (p = 0.02-0.03) were the only independent determinants of hepatocellular carcinoma microvascular invasion. In contrast to basal biomechanical parameters, the compression stiffening slope had high diagnostic performance for detecting microvascular invasion (AUCcompression stiffening = 0.83, p < 0.001). Our results suggest that the compression stiffening slope at MR elastography is useful to diagnose microvascular invasion in patients with hepatocellular carcinomas. Question Because non-invasive imaging markers of hepatocellular microvascular invasion are lacking, the development of new MRI markers is advisable. Findings In our MR elastography study, respiration-induced tumor stiffening, in contrast to basal visco-elastic parameters, had good accuracy for diagnosing hepatocellular carcinoma microvascular invasion. Clinical relevance Our results in patients with hepatocellular carcinomas suggest that the non-invasive measurement of MR elastography tumor compression stiffening slope may assess microvascular invasion.

Fractures in children are common in emergency care, and accurate diagnosis is crucial to avoid complications affecting skeletal development. Limited access to pediatric radiology specialists emphasizes the potential of artificial intelligence (AI)-based diagnostic tools. This study evaluates the performance of the AI software BoneView® for detecting fractures of the upper extremity in children aged 2-18 years. A retrospective analysis was conducted using radiographic data from 826 pediatric patients presenting to the university's pediatric emergency department. Independent assessments by two experienced pediatric radiologists served as reference standard. The diagnostic accuracy of the AI tool compared to the reference standard was evaluated and performance parameters, e.g., sensitivity, specificity, positive and negative predictive values were calculated. The AI tool achieved an overall sensitivity of 89% and specificity of 91% for detecting fractures of the upper extremities. Significantly poorer performance compared to the reference standard was observed for the shoulder, elbow, hand, and fingers, while no significant difference was found for the wrist, clavicle, upper arm, and forearm. The software performed best for wrist fractures (sensitivity: 96%; specificity: 94%) and worst for elbow fractures (sensitivity: 87%; specificity: 65%). The software assessed provides diagnostic support in pediatric emergency radiology. While its overall performance is robust, limitations in specific anatomical regions underscore the need for further training of the underlying algorithms. The results suggest that AI can complement clinical expertise but should not replace radiological assessment. Question There is no comprehensive analysis of an AI-based tool for the diagnosis of pediatric fractures focusing on the upper extremities. Findings The AI-based software demonstrated solid overall diagnostic accuracy in the detection of upper limb fractures in children, with performance differing by anatomical region. Clinical relevance AI-based fracture detection can support pediatric emergency radiology, especially where expert interpretation is limited. However, further algorithm training is needed for certain anatomical regions and for detecting associated findings such as joint effusions to maximize clinical benefit.

To assess the safety and diagnostic performance of CT-guided core needle biopsy (CNB) of focal pulmonary lesions with coexisting interstitial lung abnormalities (ILAs). This retrospective 1:1 matched case-control study included patients with ILAs and controls who underwent CT-guided CNB of a focal pulmonary lesion from February 2010 to December 2023. Complications, nondiagnostic specimens, and CNB diagnostic performance for malignancy were compared. Logistic regression was used to identify predictors of complications and nondiagnostic specimens. Resected cases were reviewed for histopathological changes in the nonneoplastic lung. Seventy-three patients with ILAs and matched controls were included (both groups: median age, 73.0 years; 13 women). No significant difference was found in complications (overall: 21/73 [29%] vs 24/73 [33%], p = 0.72; major: 2/73 [3%] vs 3/73 [4%], p = 1.00; minor: 19/73 [26%] vs 21/73 [29%], p = 0.85), nondiagnostic specimens (12/73 [16%] vs 9/73 [12%], p = 0.20), or diagnostic performance (accuracy: 89% [65/73] vs 96% [70/73], p = 0.11; sensitivity: 88% [61/69] vs 95% [62/65], p = 0.13; specificity: 100% [4/4] vs 100% [8/8], p = 1.00). Needle traversal of ILAs (OR, 7.04; 95% CI: 2.07-26.28; p = 0.008) and multiple pleural passes (OR, 8.06; 95% CI: 1.26-70.46; p = 0.03) were associated with complications. Nonneoplastic lung in ILAs revealed more complex histology and increased fibrotic features than controls. CT-guided CNB of focal pulmonary lesions with coexisting ILAs was as safe and accurate as in patients without ILAs. However, traversing ILAs and multiple pleural passes increased complication risk. Question Whether CT-guided CNB of focal pulmonary lesions with coexisting ILAs is as safe and accurate as in those without ILAs. Findings CNB in patients with ILAs showed similar safety and diagnostic performance to controls; however, complications were more frequent when ILAs were traversed or multiple pleural passes were performed. Clinical relevance CNB is a safe and effective diagnostic tool in patients with ILAs. Avoiding ILA traversal and multiple pleural passes may help minimize the risk of complications.

To provide up-to-date consensus recommendations on the acquisition, interpretation and reporting of MRI for the primary staging of rectal cancer. A panel of twenty-six abdominal imaging experts from the European Society of Gastrointestinal and Abdominal Radiology (ESGAR) engaged in an online consensus process, led by three non-voting chairs. The process adhered to an adapted version of the RAND-UCLA appropriateness method. A total of 126 items were scored (22 general, 55 on primary staging, 49 on restaging after neoadjuvant treatment) and classified using ≥ 80% as the cut-off to establish consensus. Consensus was reached for 121 items (96%). The current manuscript addresses the resulting general recommendations and those focused on baseline staging. Key updates compared to the previous guideline editions include more detailed recommendations for image acquisition, adoption of the sigmoid take-off as a landmark to discern rectal from sigmoid cancer, updated definition of mesorectal fascia involvement by a distance of ≤ 1 mm, including involvement by irregular nodes and extramural vascular invasion; a transition to a patient-level approach for cN-category assessment with updated criteria for lateral nodes including a ≥ 7 mm size threshold, and recommendations on the limited use of DWI for primary staging. These updated expert consensus recommendations serve as clinical guidelines for the primary staging of rectal cancer using MRI. Recommendations for restaging and response evaluation after neoadjuvant treatment are addressed in a separate publication. Question Since the last ESGAR rectal imaging guideline update, the rectal cancer treatment landscape has further evolved, necessitating updates to the existing guidelines. Findings An online consensus process involving 26 panellists led to 96% consensus across 121 items discussed, including 22 general items and 55 related to primary staging. Clinical relevance Key updates related to primary staging include more detailed recommendations for image acquisition, adoption of the sigmoid take-off, refined criteria for MRF involvement, a new patient-level approach for cN-assessment, and recommendations on the limited use of DWI.

To evaluate the diagnostic performance of semi-supervised learning models for aggressive prostate cancer detection on MRI compared to fully supervised models trained with additional expert annotations. We used 1500 MRI scans from the PI-CAI challenge training subset. Positive scans had 220 human and 205 AI-generated annotations. The mtU-Net (proposed teacher-student semi-supervised approach) was compared to supervised (trained using only 220 human annotations) and semi-supervised (trained on human and AI-generated annotations) nnU-Net. The 205 AI-annotated scans were manually annotated, and a fully supervised model was trained. External validation was performed on a newly annotated dataset from the PROMIS study (n = 574,403 lesions) and the Prostate158 dataset (n = 158,126 lesions). Patient-level performance was evaluated using the area under the curve (AUC) and lesion-level detection (overlap > 0.10) using average precision (AP), along with 95% confidence Intervals (in brackets), and the DeLong test to compare AUCs against the supervised and fully supervised models. The fully supervised nnU-Net showed the highest performance on the internal PI-CAI test set (AUC = 0.89 [0.87-0.91], AP = 0.65 [0.60-0.70]) and external validation datasets PROMIS (AUC = 0.68 [0.64-0.72], AP = 0.24 [0.20-0.29]) and Prostate158 (AUC = 0.87 [0.82-0.92], AP = 0.64 [0.56-0.72]), significantly outperforming the supervised baseline (p < 0.05). The proposed semi-supervised mtU-Net demonstrated close external validation performance on PROMIS (AUC = 0.66 [0.62-0.71], AP = 0.20 [0.16-0.25]) and Prostate158 (AUC = 0.86 [0.81-0.92], AP = 0.58 [0.49-0.67]), significantly outperforming the supervised baseline on both datasets (p = 0.047 and p = 0.014, respectively), and showing no significant difference to the fully supervised model (p = 0.199 and p = 0.702, respectively). In prostate MRI tumor detection, fully supervised learning performed best. However, in external validation, the semi-supervised methods demonstrated performance that approached that of the fully supervised model, proving a valuable approach when expert annotations are limited. Question The need for extensive expert voxel-level annotations delays the development of AI-based prostate cancer diagnostic tools and their implementation in clinical practice. Findings The combination of pseudo-labeling with consistency regularization achieved performance comparable to that of fully supervised methods, demonstrating that data diversity matches the impact of expert annotation volume. Clinical relevance Semi-supervised learning reduces dependence on expert annotations while maintaining detection accuracy, enabling the development of scalable, automated diagnostic tools for prostate cancer amid growing clinical workflow demands.

To develop and validate a multi-task deep learning (MTDL) model using multiphase contrast-enhanced CT (CECT) for simultaneously assessing histological subtypes, clinical stages, and anatomical complexity grades of solid malignant renal tumors. This two-center retrospective study included patients with solid malignant renal tumors and their preoperative kidney CECT images. A progressive layered extraction (PLE)-based MTDL model was trained and externally tested. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) and decision curve analysis (DCA), compared with the results of five radiologists. Among 798 patients (mean age, 54 ± 12 years; 279 females; Center A: n = 620, Center B: n = 178), 597 (74.8%) had clear cell renal cell carcinomas (ccRCC), 150 (18.8%) were clinical staging III/IV, and 187 (23.4%) had high anatomical complexity. On the external test set, the MTDL model achieved AUCs of 0.89 (95% CI: 0.82, 0.94) for distinguishing ccRCC from non-ccRCC, 0.87 (95% CI: 0.81, 0.93) for clinical staging (I/II vs. III/IV), and 0.87 (95% CI: 0.82, 0.92) for anatomical complexity grading (low-intermediate vs. high). The MTDL model outperformed single-task deep learning (STDL) in clinical staging (AUC: 0.87 vs. 0.82, p = 0.022), showed higher net benefit on DCA, and demonstrated better diagnostic performance than junior radiologists in histological subtyping and clinical staging. Additionally, it used 68% less memory and was 60% faster than STDL models. The CECT-based MTDL model demonstrated robust performance in simultaneously predicting histological subtypes, clinical stages, and anatomical complexity grades of malignant renal tumors. Question Accurate preoperative description of the histological subtyping, clinical staging, and anatomical complexity of malignant renal tumors is crucial for treatment decision-making. Findings By sharing features, the multi-task deep learning algorithm model enhances clinical staging performance and significantly improves computational efficiency in predicting all three tasks simultaneously. Clinical relevance The multi-task deep learning algorithm model enables rapid and accurate comprehensive preoperative evaluation of renal tumors, which assists surgeons in optimizing surgical plans and promotes the advancement of renal tumor management toward precision and efficiency.