An increasing number of patients with prostate cancer (PCa) undergo assessment with magnetic resonance imaging (MRI) and prostate-specific membrane antigen positron emission tomography/computed tomography (PSMA-PET/CT). This offers comprehensive multimodality staging but can lead to discrepancies. The objective was to assess the rates and types of discordance between MRI and PSMA-PET/CT for primary PCa assessment. Consecutive men diagnosed with intermediate and high-risk PCa who underwent MRI and PSMA-PET/CT in 2021–2023 were retrospectively included. MRI and PSMA-PET/CT were interpreted using PI-RADS v2.1 and PRIMARY scores. Discordances between the two imaging modalities were categorized as “minor” (larger or additional lesion seen on one modality) or “major” (positive on only one modality or different index lesions between MRI and PSMA-PET/CT) and reconciled using radical prostatectomy or biopsy specimens. Three hundred and nine men (median age 69 years, interquartile range (IQR) 64–75) were included. Most had Gleason Grade Group ≥ 3 PCa (70.9

Prostate-specific membrane antigen (PSMA)-PET/CT has become integral to management of prostate cancer; however, PSMA-avid rib lesions pose a diagnostic challenge. This study investigated clinicopathological and imaging findings that predict metastatic etiology of PSMA-avid rib lesions. Consecutive patients with prostate cancer that underwent PET/CT with [18F]F-DCFPyL in 2021–2023 for newly diagnosed intermediate-/high-risk prostate cancer or recurrent/metastatic disease and had PSMA-avid rib lesions were included. Imaging findings assessed were: lesion number, PSMA expression (maximum standard uptake value (SUVmax), miPSMA score), CT features (sclerotic, lucent, fracture, no correlate), other sites of metastases, and primary tumor findings. A composite reference standard for rib lesion etiology (metastatic vs non-metastatic) based on histopathology, serial imaging, and clinical assessment was used. One hundred and seventy-five men (median 71 years, IQR 65–77) with PSMA-avid rib lesions were included; 47/175 (26.9

Purpose To evaluate the performance of open-source, locally deployed large language models (LLMs) in anonymizing radiology reports while preserving clinical data, ensuring compliance with privacy regulations like the Health Insurance Portability and Accountability Act (HIPAA), without relying on cloud-based solutions. Methods Seven state-of-the-art open-source LLMs—Qwen-2.5-coder (7B, 32B), Llama v3.1 (8B, 70B), Llama v3.3 (70B), and Phi3/4 (14B)—were tested on 1000 randomly selected radiology reports (CT, MRI, PET/CT) from a cancer imaging cohort. Models were tasked with generating regular expression (RegEx) rules to remove protected health information (PHI) as defined by HIPAA. Inference was conducted on an A100 GPU and an M4 Max MacBook Pro using the ollama framework. Performance was assessed by three readers for protected health information (PHI) removal accuracy and preservation of non-PHI clinical data, with runtimes recorded. Results Qwen models outperformed others, with Qwen7B achieving a 97.59 % success rate in preserving clinical data (vs. 24.33 % for Llama8B) and Qwen32B achieving 100 % PHI removal (vs. 1.91 % patient names missed by Llama v3.1 70B). Llama models frequently misidentified clinical numbers as PHI, while Phi models produced unusable results due to overly aggressive rules or hallucinations. Runtimes were significantly shorter for Qwen models (e.g., Qwen7B: 2h50′ on A100 vs. Llama8B: 3h30′). Conclusion Locally deployed Qwen-2.5-coder models offer a superior, privacy-preserving solution for anonymizing radiology reports, balancing PHI removal with clinical data retention. This approach supports scalable, compliant anonymization of radiology reports for research without cloud dependency, outperforming other current state-of-the-art local LLMs.

Uncertainty in medical image segmentation tasks, especially inter-rater variability, arising from differences in interpretations and annotations by various experts, presents a significant challenge in achieving consistent and reliable image segmentation. This variability not only reflects the inherent complexity and subjective nature of medical image interpretation but also directly impacts the development and evaluation of automated segmentation algorithms. Accurately modeling and quantifying this variability is essential for enhancing the robustness and clinical applicability of these algorithms. We report the set-up and summarize the benchmark results of the Quantification of Uncertainties in Biomedical Image Quantification Challenge (QUBIQ), which was organized in conjunction with International Conferences on Medical Image Computing and Computer-Assisted Intervention (MICCAI) 2020 and 2021. The challenge focuses on the uncertainty quantification of medical image segmentation which considers the omnipresence of inter-rater variability in imaging datasets. The large collection of images with multi-rater annotations features various modalities such as MRI and CT; various organs such as the brain, prostate, kidney, and pancreas; and different image dimensions 2D-vs-3D. A total of 24 teams submitted different solutions to the problem, combining various baseline models, Bayesian neural networks, and ensemble model techniques. The obtained results indicate the importance of the ensemble models, as well as the need for further research to develop efficient 3D methods for uncertainty quantification methods in 3D segmentation tasks.

Aim: To investigate the associations between the hour of the day and Prostate Imaging-Reporting and Data System (PI-RADS) scores assigned by radiologists in prostate MRI reports. Materials and methods: Retrospective single-center collection of prostate MRI reports over an 8-year period. Mean PI-RADS scores assigned between 0800 and 1800 h were examined with a regression model. Results: A total of 35'004 prostate MRI interpretations by 26 radiologists were included. A significant association between the hour of day and mean PI-RADS score was identified (beta(2) = 0.005, p < 0.001), with malignant scores more frequently assigned later in the day. Conclusion: These findings suggest chronobiological factors may contribute to variability in radiological assessments. Though the magnitude of the effect is small, this may potentially add variability and impact diagnostic accuracy.

Prostate-specific membrane antigen (PSMA) PET/CT has an established reliable diagnostic performance for detecting metastases in prostate cancer. However, there are increasing instances of scans demonstrating equivocal bone lesions, with non-specific uptake and without a definite benign or malignant CT correlate. To date, the prevalence, malignancy rate, and relationship with radioligand type ([18F] PSMA-1007 vs. others ([68Ga]Ga-PSMA-11 and [18F] DCFPyL) for these equivocal lesions have not been extensively established. A systematic review and meta-analysis was conducted on equivocal bone lesions. Pubmed and EMBASE were searched up to December 11, 2023. Quality of the studies was evaluated using QUADAS-2. The following proportions were pooled using random-effects model: (1) prevalence of equivocal bone lesions (i.e., number of patients with one or more equivocal bone lesions/number of patients with PSMA PET/CT) and (2) their malignancy rates (i.e., number of metastases/number of equivocal bone lesions). Subgroup analyses based on radioligand type, clinical setting, and definition of equivocal bone lesion were performed. Twenty-five studies (4484 patients) were included. Pooled prevalence of equivocal bone lesions was 20

Background The rising global cancer burden has led to an increasing demand for imaging tests such as [F-18]fluorodeoxyglucose ([F-18]FDG)-PET-CT. To aid imaging specialists in dealing with high scan volumes, we aimed to train a deep learning artificial intelligence algorithm to classify [F-18]FDG-PET-CT scans of patients with lymphoma with or without hypermetabolic tumour sites. Methods In this retrospective analysis we collected 16 583 [F-18]FDG-PET-CTs of 5072 patients with lymphoma who had undergone PET-CT before or after treatment at the Memorial Sloa Kettering Cancer Center, New York, NY, USA. Using maximum intensity projection (MIP), three dimensional (3D) PET, and 3D CT data, our ResNet34-based deep learning model (Lymphoma Artificial Reader System [LARS]) for [F-18]FDG-PET-CT binary classification (Deauville 1-3 vs 4-5), was trained on 80% of the dataset, and tested on 20% of this dataset. For external testing, 1000 [F-18]FDG-PET-CTs were obtained from a second centre (Medical University of Vienna, Vienna, Austria). Seven model variants were evaluated, including MIP-based LARS-avg (optimised for accuracy) and LARS-max (optimised for sensitivity), and 3D PET-CT-based LARS-ptct. Following expert curation, areas under the curve (AUCs), accuracies, sensitivities, and specificities were calculated. Findings In the internal test cohort (3325 PET-CTs, 1012 patients), LARS-avg achieved an AUC of 0.949 (95% CI 0.942-0.956), accuracy of 0.890 (0.879-0.901), sensitivity of 0.868 (0.851-0.885), and specificity of 0.913 (0.899-0.925); LARS-max achieved an AUC of 0.949 (0.942-0.956), accuracy of 0.868 (0.858-0.879), sensitivity of 0.909 (0.896-0.924), and specificity of 0.826 (0.808-0.843); and LARS-ptct achieved an AUC of 0.939 (0.930-0.948), accuracy of 0.875 (0.864-0.887), sensitivity of 0.836 (0.817-0.855), and specificity of 0.915 (0.901-0.927). In the external test cohort (1000 PET-CTs, 503 patients), LARS-avg achieved an AUC of 0.953 (0.938-0.966), accuracy of 0.907 (0.888-0.925), sensitivity of 0.874 (0.843-0.904), and specificity of 0.949 (0.921-0.960); LARS-max achieved an AUC of 0.952 (0.937-0.965), accuracy of 0.898 (0.878-0.916), sensitivity of 0.899 (0.871-0.926), and specificity of 0.897 (0.871-0.922); and LARS-ptct achieved an AUC of 0.932 (0.915-0.948), accuracy of 0.870 (0.850-0.891), sensitivity of 0.827 (0.793-0.863), and specificity of 0.913 (0.889-0.937). Interpretation Deep learning accurately distinguishes between [F-18]FDG-PET-CT scans of lymphoma patients with and without hypermetabolic tumour sites. Deep learning might therefore be potentially useful to rule out the presence of metabolically active disease in such patients, or serve as a second reader or decision support tool. Copyright (c) 2023 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY 4.0 license.

The standard method for identifying active Brown Adipose Tissue (BAT) is [ 18 F]-Fluorodeoxyglucose ([ 18 F]-FDG) PET/CT imaging, which is costly and exposes patients to radiation, making it impractical for population studies. These issues can be addressed with computational methods that predict [ 18 F]-FDG uptake by BAT from CT; earlier population studies pave the way for developing such methods by showing some correlation between the Hounsfield Unit (HU) of BAT in CT and the corresponding [ 18 F]-FDG uptake in PET. In this study, we propose training convolutional neural networks (CNNs) to predict [ 18 F]-FDG uptake by BAT from unenhanced CT scans in the restricted regions which are likely to contain BAT. We performed experiments on datasets from four different cohorts, the largest study to date. We segmented the BAT regions using the predicted [ 18 F]-FDG uptake values and evaluated the accuracy of the segmentations, showing 23% to 40% improvement over the conventional CT thresholding-based method. Additionally, we showed that BAT volumes computed from the segmentations could be used to distinguish subjects with and without active BAT with an AUC of 0.8, whereas the CT thresholding-based method achieved 0.6. Our findings suggest that CNNs, when trained on cold-stimulated cohorts, can be effectively used to create stratified cohorts from only CT, thereby having the potential for facilitating large-scale imaging studies with greater efficiency and lower costs.

Purpose: To describe the structure of a dedicated body oncologic imaging fellowship program. To summarize the numbers and types of cross-sectional imaging examinations reported by fellows. Methods: The curriculum, training methods, and assessment measures utilized in the program were reviewed and described. An educational retrospective analysis was conducted. Data on the number of examinations interpreted by fellows, breakdown of modalities, and examinations by disease management team (DMT) were collected. Results: A total of 38 fellows completed the fellowship program during the study period. The median number of examinations reported per fellow was 2296 [interquartile range: 2148 - 2534], encompassing all oncologyrelevant imaging modalities: CT 721 [646-786], MRI 1158 [1016-1309], ultrasound 256 [209-320] and PET/CT 176 [130-202]. The breakdown of examinations by DMT revealed variations in imaging patterns, with MRIs most frequently interpreted for genitourinary, musculoskeletal, and hepatobiliary cancers, and CTs most commonly for general staging or assessment of nonspecific symptoms. Conclusion: This descriptive analysis may serve as a foundation for the development of similar fellowship programs and the advancement of body oncologic imaging. The volume and diversity of examinations reported by fellows highlights the comprehensive nature of body oncologic imaging.

Aims:Treatment of high-grade limb bone sarcoma that invades a joint requires en bloc extra-articular excision. MRI can demonstrate joint invasion but is frequently inconclusive, and its predictive value is unknown. We evaluated the diagnostic accuracy of direct and indirect radiological signs of intra-articular tumour extension and the performance characteristics of MRI findings of intra-articular tumour extension. Methods:We performed a retrospective case-control study of patients who underwent extra-articular excision for sarcoma of the knee, hip, or shoulder from 1 June 2000 to 1 November 2020. Radiologists blinded to the pathology results evaluated preoperative MRI for three direct signs of joint invasion (capsular disruption, cortical breach, cartilage invasion) and indirect signs (e.g. joint effusion, synovial thickening). The discriminatory ability of MRI to detect intra-articular tumour extension was determined by receiver operating characteristic analysis. Results:Overall, 49 patients underwent extra-articular excision. The area under the curve (AUC) ranged from 0.65 to 0.76 for direct signs of joint invasion, and was 0.83 for all three combined. In all, 26 patients had only one to two direct signs of invasion, representing an equivocal result. In these patients, the AUC was 0.63 for joint effusion and 0.85 for synovial thickening. When direct signs and synovial thickening were combined, the AUC was 0.89. Conclusion:MRI provides excellent discrimination for determining intra-articular tumour extension when multiple direct signs of invasion are present. When MRI results are equivocal, assessment of synovial thickening increases MRI's discriminatory ability to predict intra-articular joint extension. These results should be interpreted in the context of the study's limitations. The inclusion of only extra-articular excisions enriched the sample for true positive cases. Direct signs likely varied with tumour histology and location. A larger, prospective study of periarticular bone sarcomas with spatial correlation of histological and radiological findings is needed to validate these results before their adoption in clinical practice.

Microservices are a software development approach where an application is structured as a collection of loosely coupled, independently deployable services, each focusing on executing a specific purpose. The development of microservices could have a significant impact on radiology workflows, allowing routine tasks to be automated and improving the efficiency and accuracy of radiologic tasks. This technical report describes the development of several microservices that have been successfully deployed in a tertiary cancer center, resulting in substantial time savings for radiologists and other staff involved in radiology workflows. These microservices include the automatic generation of shift emails, notifying administrative staff and faculty about fellows on rotation, notifying referring physicians about outside examinations, and populating report templates with information from PACS and RIS. The report outlines the common thought process behind developing these microservices, including identifying a problem, connecting various APIs, collecting data in a database, writing a prototype and deploying it, gathering feedback and refining the service, putting it in production, and identifying staff who are in charge of maintaining the service. The report concludes by discussing the benefits and challenges of microservices in radiology workflows, highlighting the importance of multidisciplinary collaboration, interoperability, security, and privacy.

Objective:To investigate clinical, pathology, and imaging findings associated with inguinal lymph node (LN) metastases in patients with prostate cancer (PCa). Materials and Methods:This was a retrospective single-center study of patients with PCa who underwent imaging and inguinal LN biopsy between 2000 and 2023. We assessed the following aspects on multimodality imaging: inguinal LN morphology; extrainguinal lymphadenopathy; the extent of primary and recurrent tumors; and non-nodal metastases. Imaging, clinical, and pathology features were compared between patients with and without metastatic inguinal LNs. Results:We evaluated 79 patients, of whom 38 (48.1%) had pathology-proven inguinal LN metastasis. Certain imaging aspects- short-axis diameter, prostate-specific membrane antigen uptake on positron-emission tomography, membranous urethra involvement by the tumor, extra-inguinal lymphadenopathy, and distant metastases-were associated with pathology-proven inguinal LN metastases (p < 0.01 for all). Associations with long-axis diameter, fatty hilum, laterality, and uptake of other tracers on positronemission tomography were not significant (p = 0.09-1.00). The patients with metastatic inguinal LNs had higher prostate-specific antigen levels and more commonly had castration-resistant PCa (p < 0.01), whereas age, histological grade, and treatment type were not significant factors (p = 0.07-0.37). None of the patients had inguinal LN metastasis in the absence of locally advanced disease with membranous urethra involvement or distant metastasis. Conclusion:Several imaging, clinical, and pathology features are associated with inguinal LN metastases in patients with PCa. Isolated metastasis to inguinal LNs is extremely rare and unlikely to occur in the absence of high-risk imaging, clinical, or pathology features.

ObjectiveTo evaluate MRI features of sarcomatoid renal cell carcinoma (RCC) and their association with survival.MethodsThis retrospective single-center study included 59 patients with sarcomatoid RCC who underwent MRI before nephrectomy during July 2003-December 2019. Three radiologists reviewed MRI findings of tumor size, non-enhancing areas, lymphadenopathy, and volume (and percentage) of T2 low signal intensity areas (T2LIA). Clinicopathological factors of age, gender, ethnicity, baseline metastatic status, pathological details (subtype and extent of sarcomatoid differentiation), treatment type, and follow-up were extracted. Survival was estimated using Kaplan-Meier method and Cox proportional-hazards regression model was used to identify factors associated with survival.ResultsForty-one males and eighteen females (median age 62 years; interquartile range 51-68) were included. T2LIAs were present in 43 (72.9%) patients. At univariate analysis, clinicopathological factors associated with shorter survival were: greater tumor size (> 10 cm; HR [hazard ratio] = 2.44, 95% CI 1.15-5.21; p = 0.02), metastatic lymph nodes (present; HR = 2.10, 95% CI 1.01-4.37; p = 0.04), extent of sarcomatoid differentiation (non-focal; HR = 3.30, 95% CI 1.55-7.01; p < 0.01), subtypes other than clear cell, papillary, or chromophobe (HR = 3.25, 95% CI 1.28-8.20; p = 0.01), and metastasis at baseline (HR = 5.04, 95% CI 2.40-10.59; p < 0.01). MRI features associated with shorter survival were: lymphadenopathy (HR = 2.24, 95% CI 1.16-4.71; p = 0.01) and volume of T2LIA (> 3.2 mL, HR = 4.22, 95% CI 1.92-9.29); p < 0.01). At multivariate analysis, metastatic disease (HR = 6.89, 95% CI 2.79-16.97; p < 0.01), other subtypes (HR = 9.50, 95% CI 2.81-32.13; p < 0.01), and greater volume of T2LIA (HR = 2.51, 95% CI 1.04-6.05; p = 0.04) remained independently associated with worse survival.ConclusionT2LIAs were present in approximately two thirds of sarcomatoid RCCs. Volume of T2LIA along with clinicopathological factors were associated with survival.

Background Neoadjuvant chemotherapy (NAC) before radical cystectomy is standard of care in patients with muscle-invasive bladder cancer (MIBC). Response assessment after NAC is important but suboptimal using CT. We assessed MRI without vs. with intravenous contrast (biparametric [BP] vs. multiparametric [MP]) for identifying residual disease on cystectomy and explored its prognostic role. Methods Consecutive MIBC patients that underwent NAC, MRI, and cystectomy between January 2000–November 2022 were identified. Two radiologists reviewed BP-MRI (T2 + DWI) and MP-MRI (T2 + DWI + DCE) for residual tumor. Diagnostic performances were compared using receiver operating characteristic curve analysis. Kaplan-Meier curves and Cox proportional-hazards models were used to evaluate association with disease-free survival (DFS). Results 61 patients (36 men and 25 women; median age 65 years, interquartile range 59–72) were included. After NAC, no residual disease was detected on pathology in 19 (31.1%) patients. BP-MRI was more accurate than MP-MRI for detecting residual disease after NAC: area under the curve = 0.75 (95% confidence interval (CI), 0.62–0.85) vs. 0.58 (95% CI, 0.45–0.70; p = 0.043). Sensitivity were identical (65.1%; 95% CI, 49.1–79.0) but specificity was higher in BP-MRI compared with MP-MRI for determining residual disease: 77.8% (95% CI, 52.4–93.6) vs. 38.9% (95% CI, 17.3–64.3), respectively. Positive BP-MRI and residual disease on pathology were both associated with worse DFS: hazard ratio (HR) = 4.01 (95% CI, 1.70–9.46; p = 0.002) and HR = 5.13 (95% CI, 2.66–17.13; p = 0.008), respectively. Concordance between MRI and pathology results was significantly associated with DFS. Concordant positive (MRI+/pathology+) patients showed worse DFS than concordant negative (MRI-/pathology-) patients (HR = 8.75, 95% CI, 2.02–37.82; p = 0.004) and compared to the discordant group (MRI+/pathology- or MRI-/pathology+) with HR = 3.48 (95% CI, 1.39–8.71; p = 0.014). Conclusion BP-MRI was more accurate than MP-MRI for identifying residual disease after NAC. A negative BP-MRI was associated with better outcomes, providing complementary information to pathological assessment of cystectomy specimens.

PurposeTo compare the interreader agreement of a novel quality score, called the Radiological Image Quality Score (RI-QUAL), to a slighly modified version of the existing Prostate Imaging Quality (mPI-QUAL) score for magnetic resonance imaging (MRI) of the prostate.MethodsA total of 43 consecutive scans were evaluated by two subspecialized radiologists who assigned scores using both the RI-QUAL and mPI-QUAL methods. The interreader agreement was analyzed using three statistical methods: concordance correlation coefficient (CCC), intraclass correlation coefficient (ICC), and Cohen's kappa. Time needed to arrive at a quality judgment was measured and compared using the Wilcoxon signed rank test.ResultsThe interreader agreement for RI-QUAL and mPI-QUAL scores was comparable, as evidenced by the high CCC (0.76 vs. 0.77, p = 0.93), ICC (0.86 vs. 0.87, p = 0.93), and moderate Cohen's kappa (0.61 vs. 0.64, p = 0.85) values. Moreover, RI-QUAL assessment was faster than mPI-QUAL (19 vs. 40 s, p = 0.001).ConclusionRI-QUAL is a new quality score that has comparable interreader agreement to the mPI-QUAL score, but with the potential to be applied to different MRI protocols and even different modalities. Like PI-QUAL, RI-QUAL may also facilitate communication about quality to referring physicians, as it provides a standardized and easily interpretable score. Further studies are warranted to validate the usefulness of RI-QUAL in larger patient cohorts and for other imaging modalities.

In a retrospective single-center study, the authors assessed the efficacy of an automated imaging examination assignment system for enhancing the diversity of subspecialty examinations reported by oncologic imaging fellows. The study aimed to mitigate traditional biases of manual case selection and ensure equitable exposure to various case types. Methods included evaluating the proportion of "uncommon" to "common" cases reported by fellows before and after system implementation and measuring the weekly Shannon Diversity Index to determine case distribution equity. The proportion of reported uncommon cases more than doubled from 8.6% to 17.7% in total, at the cost of a concurrent 9.0% decrease in common cases from 91.3% to 82.3%. The weekly Shannon Diversity Index per fellow increased significantly from 0.66 (95% CI: 0.65, 0.67) to 0.74 (95% CI: 0.72, 0.75; P < .001), confirming a more balanced case distribution among fellows after introduction of the automatic assignment. © RSNA, 2023 Keywords: Computer Applications, Education, Fellows, Informatics, MRI, Oncologic Imaging.

Lung cancer is the leading cause of cancer death worldwide, with lung adenocarcinoma being the most prevalent form of lung cancer. EGFR positive lung adenocarcinomas have been shown to have high response rates to TKI therapy, underlying the essential nature of molecular testing for lung cancers. Despite current guidelines consider testing necessary, a large portion of patients are not routinely profiled, resulting in millions of people not receiving the optimal treatment for their lung cancer. Sequencing is the gold standard for molecular testing of EGFR mutations, but it can take several weeks for results to come back, which is not ideal in a time constrained scenario. The development of alternative screening tools capable of detecting EGFR mutations quickly and cheaply while preserving tissue for sequencing could help reduce the amount of sub-optimally treated patients. We propose a multi-modal approach which integrates pathology images and clinical variables to predict EGFR mutational status achieving an AUC of 84% on the largest clinical cohort to date. Such a computational model could be deployed at large at little additional cost. Its clinical application could reduce the number of patients who receive sub-optimal treatments by 53.1% in China, and up to 96.6% in the US.

In patients with symptomatic ureterolithiasis, immediate treatment of concomitant urinary tract infection (UTI) may prevent sepsis. However, urine cultures require at least 24 h to confirm or exclude UTI, and therefore, clinical variables may help to identify patients who require immediate empirical broad-spectrum antibiotics and surgical intervention. Therefore, we divided a consecutive cohort of 705 patients diagnosed with symptomatic ureterolithiasis at a single institution between 2011 and 2017 into a training (80%) and a testing cohort (20%). A machine-learning-based variable selection approach was used for the fitting of a multivariable prognostic logistic regression model. The discriminatory ability of the model was quantified by the area under the curve (AUC) of receiver-operating curves (ROC). After validation and calibration of the model, a nomogram was created, and decision curve analysis (DCA) was used to evaluate the clinical net-benefit. UTI was observed in 40 patients (6%). LASSO regression selected the variables elevated serum CRP, positive nitrite, and positive leukocyte esterase for fitting of the model with the highest discriminatory ability. In the testing cohort, model performance evaluation for prediction of UTI showed an AUC of 82 (95% CI 71.5-95.7%). Model calibration plots showed excellent calibration. DCA showed a clinically meaningful net-benefit between a threshold probability of 0 and 80% for the novel model, which was superior to the net-benefit provided by either one of its singular components. In conclusion, we developed and internally validated a logistic regression model and a corresponding highly accurate nomogram for prediction of concomitant positive midstream urine culture in patients presenting with symptomatic ureterolithiasis.

With the increasing development of artificial intelligence (AI) and deep learning (DL) technology, the application of DL networks in medical image analysis has attracted more and more attention in recent years. Hence, publications in our field amalgamating DL and radiological problems have soared exponentially [[1]Saba L. Biswas M. Kuppili V. Cuadrado Godia E. Suri H.S. Edla D.R. Omerzu T. Laird J.R. Khanna N.N. Mavrogeni S. Protogerou A. Sfikakis P.P. Viswanathan V. Kitas G.D. Nicolaides A. Gupta A. Suri J.S. The present and future of deep learning in radiology.Eur. J. Radiol. 2019; 114: 14-24https://doi.org/10.1016/j.ejrad.2019.02.038Abstract Full Text Full Text PDF PubMed Scopus (154) Google Scholar]. Even for someone actively working in the field it becomes increasingly difficult to keep a high-level overview across the general trends.In their comprehensive review, Wang et al. [[2]Wang L.u. Wang H. Huang Y. Yan B. Chang Z. Liu Z. Zhao M. Cui L. Song J. Li F. Trends in the application of deep learning networks in medical image analysis: evolution between 2012 and 2020.Eur. J. Radiol. 2022; 146: 110069https://doi.org/10.1016/j.ejrad.2021.110069Abstract Full Text Full Text PDF PubMed Scopus (10) Google Scholar] conducted a systematic analysis of 2685 original articles published in PubMed to summarize the general rules of application of DL networks to medical image analysis. Their results, some of which are summarized below, showed that studies related to neuroradiology, thorax, and abdomen were the most popular subspecialties (40.7%), while studies related to thyroid and dermatology were under-reported (3.9%).In 2020, thorax-related studies exceeded other studies owing to the COVID-19 pandemic, which comprised a staggering 18.7% of 2020 studies. As shown in Fig. 1, this came at the cost of decreased publication volume in nearly all other fields. Wynants et al. found no clinical benefit of the majority of clinical modeling studies pertaining to COVID-19 [[3]Wynants L. Van Calster B. Collins G.S. Riley R.D. Heinze G. Schuit E. et al.Prediction models for diagnosis and prognosis of covid-19: systematic review and critical appraisal.BMJ. 2020; 369: m1328https://doi.org/10.1136/bmj.m1328Crossref PubMed Scopus (1369) Google Scholar]. Hopefully, the insights from the radiological AI studies will be transferable to other fields where DL remains under-developed, such as thyroid imaging – and/or help to mitigate the next pandemic.Based on MeSH terms clustering, apart from lung studies (nearly half of which were COVID-19 related in 2020), brain- and breast-related diseases were research hot spots. Interestingly, brain-, prostate- and diabetic-retinopathy-related studies were considered more developed research topics than e.g. breast or lung topics, despite some large-scale studies in those fields, e.g. [[4]McKinney S.M. Sieniek M. Godbole V. Godwin J. Antropova N. Ashrafian H. et al.International evaluation of an AI system for breast cancer screening.Nature. 2020; 577: 89-94https://doi.org/10.1038/s41586-019-1799-6Crossref PubMed Scopus (765) Google Scholar]. The authors attribute this to the better availability of large public datasets. This highlights the importance of data sharing as a catalyst for research progress and, ultimately, clinical adoption.Compared with other existing reviews in the field, this study also evaluated each of the current DL networks and their application in various human organs and radiological tasks. Interestingly, the top 9 networks accounted for 94.1% of articles, highlighting again how making data/code easily available can fuel adoption and further research.With more novel algorithms based on generative adversarial nets (GAN), the authors expect an ongoing new trend in image generation and enhancement, as well as semantic feature extraction. Indeed, recent studies have explored tasks like bone-suppression [[5]Zhou Z. Zhou L. Shen K. Dilated conditional GAN for bone suppression in chest radiographs with enforced semantic features.Med. Phys. 2020; 47: 6207-6215https://doi.org/10.1002/mp.14371Crossref PubMed Scopus (6) Google Scholar], MR relaxation map generation [[6]Sveinsson B. Chaudhari A.S. Zhu B.o. Koonjoo N. Torriani M. Gold G.E. Rosen M.S. Synthesizing quantitative T2 maps in right lateral knee femoral condyles from multicontrast anatomic data with a conditional generative adversarial network.Radiol. Artif. Intell. 2021; 3: e200122https://doi.org/10.1148/ryai.2021200122Crossref PubMed Scopus (4) Google Scholar] or even cyber-attacks [7Becker A.S. Jendele L. Skopek O. Berger N. Ghafoor S. Marcon M. Konukoglu E. Injecting and removing suspicious features in breast imaging with CycleGAN: A pilot study of automated adversarial attacks using neural networks on small images.Eur. J. Radiol. 2019; 120: 108649https://doi.org/10.1016/j.ejrad.2019.108649Abstract Full Text Full Text PDF PubMed Scopus (12) Google Scholar, 8Zhou Q. Zuley M. Guo Y. Yang L.u. Nair B. Vargo A. Ghannam S. Arefan D. Wu S. A machine and human reader study on AI diagnosis model safety under attacks of adversarial images.Nat. Commun. 2021; 12https://doi.org/10.1038/s41467-021-27577-xCrossref Scopus (3) Google Scholar] powered by GAN.Lastly, the proportion of first authors with a non-clinical background was 56.4%. Therefore, appropriate clinical training of researchers in this field and collaborations between clinicians and computational scientists should be actively fostered.At EJR, we will generally favor submissions pertaining to concrete clinical use cases [[9]Nensa F. Editorial comment to artificial intelligence X-ray measurement technology of anatomical parameters related to lumbosacral stability.Eur. J. Radiol. 2022; 148: 110143https://doi.org/10.1016/j.ejrad.2021.110143Abstract Full Text Full Text PDF PubMed Scopus (1) Google Scholar] and using state-of-the-art technology (e.g., DL as opposed to texture analysis [10Pinto dos Santos D. Radiomics in endometrial cancer and beyond - a perspective from the editors of the EJR.Eur. J. Radiol. 2022; 150: 110266https://doi.org/10.1016/j.ejrad.2022.110266Abstract Full Text Full Text PDF PubMed Scopus (2) Google Scholar, 11Summers R.M. Texture analysis in radiology: does the emperor have no clothes?.Abdom. Radiol. (NY). 2017; 42: 342-345https://doi.org/10.1007/s00261-016-0950-1Crossref PubMed Scopus (42) Google Scholar, 12Moskowitz C.S. Welch M.L. Jacobs M.A. Kurland B.F. Simpson A.L. Radiomic analysis: study design, statistical analysis, and other bias mitigation strategies.Radiology. 2022; 304: 265-273https://doi.org/10.1148/radiol.211597Crossref PubMed Scopus (3) Google Scholar]). However, open data as well as active collaboration between researchers and clinicians are key factors for progress and adoption of DL research. Hence, code sharing and open data are additional important considerations in our editorial and peer review process. With the increasing development of artificial intelligence (AI) and deep learning (DL) technology, the application of DL networks in medical image analysis has attracted more and more attention in recent years. Hence, publications in our field amalgamating DL and radiological problems have soared exponentially [[1]Saba L. Biswas M. Kuppili V. Cuadrado Godia E. Suri H.S. Edla D.R. Omerzu T. Laird J.R. Khanna N.N. Mavrogeni S. Protogerou A. Sfikakis P.P. Viswanathan V. Kitas G.D. Nicolaides A. Gupta A. Suri J.S. The present and future of deep learning in radiology.Eur. J. Radiol. 2019; 114: 14-24https://doi.org/10.1016/j.ejrad.2019.02.038Abstract Full Text Full Text PDF PubMed Scopus (154) Google Scholar]. Even for someone actively working in the field it becomes increasingly difficult to keep a high-level overview across the general trends. In their comprehensive review, Wang et al. [[2]Wang L.u. Wang H. Huang Y. Yan B. Chang Z. Liu Z. Zhao M. Cui L. Song J. Li F. Trends in the application of deep learning networks in medical image analysis: evolution between 2012 and 2020.Eur. J. Radiol. 2022; 146: 110069https://doi.org/10.1016/j.ejrad.2021.110069Abstract Full Text Full Text PDF PubMed Scopus (10) Google Scholar] conducted a systematic analysis of 2685 original articles published in PubMed to summarize the general rules of application of DL networks to medical image analysis. Their results, some of which are summarized below, showed that studies related to neuroradiology, thorax, and abdomen were the most popular subspecialties (40.7%), while studies related to thyroid and dermatology were under-reported (3.9%). In 2020, thorax-related studies exceeded other studies owing to the COVID-19 pandemic, which comprised a staggering 18.7% of 2020 studies. As shown in Fig. 1, this came at the cost of decreased publication volume in nearly all other fields. Wynants et al. found no clinical benefit of the majority of clinical modeling studies pertaining to COVID-19 [[3]Wynants L. Van Calster B. Collins G.S. Riley R.D. Heinze G. Schuit E. et al.Prediction models for diagnosis and prognosis of covid-19: systematic review and critical appraisal.BMJ. 2020; 369: m1328https://doi.org/10.1136/bmj.m1328Crossref PubMed Scopus (1369) Google Scholar]. Hopefully, the insights from the radiological AI studies will be transferable to other fields where DL remains under-developed, such as thyroid imaging – and/or help to mitigate the next pandemic. Based on MeSH terms clustering, apart from lung studies (nearly half of which were COVID-19 related in 2020), brain- and breast-related diseases were research hot spots. Interestingly, brain-, prostate- and diabetic-retinopathy-related studies were considered more developed research topics than e.g. breast or lung topics, despite some large-scale studies in those fields, e.g. [[4]McKinney S.M. Sieniek M. Godbole V. Godwin J. Antropova N. Ashrafian H. et al.International evaluation of an AI system for breast cancer screening.Nature. 2020; 577: 89-94https://doi.org/10.1038/s41586-019-1799-6Crossref PubMed Scopus (765) Google Scholar]. The authors attribute this to the better availability of large public datasets. This highlights the importance of data sharing as a catalyst for research progress and, ultimately, clinical adoption. Compared with other existing reviews in the field, this study also evaluated each of the current DL networks and their application in various human organs and radiological tasks. Interestingly, the top 9 networks accounted for 94.1% of articles, highlighting again how making data/code easily available can fuel adoption and further research. With more novel algorithms based on generative adversarial nets (GAN), the authors expect an ongoing new trend in image generation and enhancement, as well as semantic feature extraction. Indeed, recent studies have explored tasks like bone-suppression [[5]Zhou Z. Zhou L. Shen K. Dilated conditional GAN for bone suppression in chest radiographs with enforced semantic features.Med. Phys. 2020; 47: 6207-6215https://doi.org/10.1002/mp.14371Crossref PubMed Scopus (6) Google Scholar], MR relaxation map generation [[6]Sveinsson B. Chaudhari A.S. Zhu B.o. Koonjoo N. Torriani M. Gold G.E. Rosen M.S. Synthesizing quantitative T2 maps in right lateral knee femoral condyles from multicontrast anatomic data with a conditional generative adversarial network.Radiol. Artif. Intell. 2021; 3: e200122https://doi.org/10.1148/ryai.2021200122Crossref PubMed Scopus (4) Google Scholar] or even cyber-attacks [7Becker A.S. Jendele L. Skopek O. Berger N. Ghafoor S. Marcon M. Konukoglu E. Injecting and removing suspicious features in breast imaging with CycleGAN: A pilot study of automated adversarial attacks using neural networks on small images.Eur. J. Radiol. 2019; 120: 108649https://doi.org/10.1016/j.ejrad.2019.108649Abstract Full Text Full Text PDF PubMed Scopus (12) Google Scholar, 8Zhou Q. Zuley M. Guo Y. Yang L.u. Nair B. Vargo A. Ghannam S. Arefan D. Wu S. A machine and human reader study on AI diagnosis model safety under attacks of adversarial images.Nat. Commun. 2021; 12https://doi.org/10.1038/s41467-021-27577-xCrossref Scopus (3) Google Scholar] powered by GAN. Lastly, the proportion of first authors with a non-clinical background was 56.4%. Therefore, appropriate clinical training of researchers in this field and collaborations between clinicians and computational scientists should be actively fostered. At EJR, we will generally favor submissions pertaining to concrete clinical use cases [[9]Nensa F. Editorial comment to artificial intelligence X-ray measurement technology of anatomical parameters related to lumbosacral stability.Eur. J. Radiol. 2022; 148: 110143https://doi.org/10.1016/j.ejrad.2021.110143Abstract Full Text Full Text PDF PubMed Scopus (1) Google Scholar] and using state-of-the-art technology (e.g., DL as opposed to texture analysis [10Pinto dos Santos D. Radiomics in endometrial cancer and beyond - a perspective from the editors of the EJR.Eur. J. Radiol. 2022; 150: 110266https://doi.org/10.1016/j.ejrad.2022.110266Abstract Full Text Full Text PDF PubMed Scopus (2) Google Scholar, 11Summers R.M. Texture analysis in radiology: does the emperor have no clothes?.Abdom. Radiol. (NY). 2017; 42: 342-345https://doi.org/10.1007/s00261-016-0950-1Crossref PubMed Scopus (42) Google Scholar, 12Moskowitz C.S. Welch M.L. Jacobs M.A. Kurland B.F. Simpson A.L. Radiomic analysis: study design, statistical analysis, and other bias mitigation strategies.Radiology. 2022; 304: 265-273https://doi.org/10.1148/radiol.211597Crossref PubMed Scopus (3) Google Scholar]). However, open data as well as active collaboration between researchers and clinicians are key factors for progress and adoption of DL research. Hence, code sharing and open data are additional important considerations in our editorial and peer review process. The author declares that he has no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Anton Becker, MD, PhD, is a faculty member at the department of radiology, Memorial Sloan Kettering Cancer Center and assistant professor of radiology at Weill Cornell Medical College in New York City (USA). He is subspecialized in imaging of genitourinary malignancies and sarcomas, and serves as the director for analytics in the body imaging service. His research interests are the application of advanced statistical computing and machine learning methods in cancer imaging.