Multi-sequence MRI protocols for glioma classification are resource-intensive, prompting the need to identify optimal, potentially reduced imaging protocols. To evaluate the diagnostic utility of individual MRI sequences (ASL, DWI, FLAIR, SWI, T1w, T2w) for classifying WHO CNS grade (II, III, IV), final pathological diagnosis (glioblastoma vs. astrocytoma), and IDH status (mutated, wildtype, IDH1) using radiomic features and machine learning. We analyzed data from 501 patients with diagnosed brain cancer from the UCSF-PDGM dataset, by extracting radiomics features from all available imaging sequences and training uni- and multi-sequence machine learning models (ASL, DWI, FLAIR, SWI, T1w, T2w) for the three outcome tasks of interest. Performance metrics (accuracy, sensitivity, F1-score) were derived for each task, alongside feature selection frequencies. The uni-sequence machine learning models using FLAIR imaging features excelled for WHO CNS grade classification (57.4
To compare the conspicuity of melanoma brain metastases on post-contrast 3D T1 gradient-recalled echo (GRE) and 3D T1 turbo spin-echo (TSE) black blood sequences, according to metastases composition. In this prospective single-center study, consecutive adult melanoma patients underwent contrast-enhanced brain MRI at 1.5 T, including standardized post-contrast 3D T1 GRE and 3D T1 TSE black blood sequences. Metastases were categorized as melanotic, non-melanotic, or hemorrhagic based on pre-contrast T1WI and susceptibility-weighted imaging (SWI). Two blinded neuroradiologists independently rated lesion conspicuity using a 5-point comparative visual scale. Interobserver agreement was assessed using the quadratically weighted Cohen’s kappa coefficient. For lesions ≥ 5 mm visible on both sequences, contrast-to-noise ratio (CNR) was measured and compared using paired non-parametric tests. A total of 261 brain metastases in 27 patients were analyzed (melanotic n = 121, non-melanotic n = 114, and hemorrhagic n = 26). Melanotic metastases were more conspicuous on post-contrast 3D T1 GRE, with 70.3
The purpose of this study was to develop a machine learning model based on radiomic features extracted from baseline contrast-enhanced CT scans for predicting early treatment responses in two cohorts of patients with stage III-IV NSCLC treated with distinct therapeutic regimens (first-line chemo-immunotherapy or immunotherapy alone). In this retrospective bicentric study including two cohorts, patients with a confirmed diagnosis of advanced NSCLC (stage III-IV) were retrospectively collected. At Center 1, after application of the exclusion criteria, 90 eligible patients (52 treated with CHT/IT and 38 with IT) were included for model development and internal validation. An independent external cohort from Center 2 comprised 40 additional patients, including 20 treated with CHT/IT and 20 with IT, and was used for external validation. Radiomic features were extracted from segmented tumor volumes using the Trace4Research™ platform, and multiple machine learning models, including random forest, support vector machine, K-nearest neighbors, multilayer perceptron, and logistic regression, were trained and compared. In the CHT/IT cohort, seven radiomic features were retained after outlier removal. Among the tested models, the support vector machine (SVM) classifier showed the best performance, with a ROC-AUC of 0.90 (95
Compare the radiological characteristics of screening images preceding interval cancers that occurred in women screened with digital breast tomosynthesis (DBT) plus digital mammography (DM) vs. those in women screened with DM alone. From two randomized trials comparing DBT + DM vs. DM, 91 (43 from DBT + DM arm and 48 from DM arm) images preceding interval cancers and 190 (94 and 96, respectively) negative controls were reviewed by three radiologists using only DM screening mammograms and by three different radiologists using all available images, i.e. DBT and DM for the experimental arm and DM for the standard arm. The cancers were classified according to the number of reviewers that found abnormalities, as true negative (negative for all three reviewers), minimal sign (positive for one and negative for two), and false negative (positive for two or three reviewers). In the DM arm, interval cancers were classified as true negative in 29–42
Accurately assessing tumor response to immunotherapy remains a significant clinical challenge due to the unique response patterns and limitations of conventional anatomical imaging. This review explores the evolving application of PET/CT in evaluating immunotherapy efficacy through the novel lens of tumor metabolic reprogramming and immune escape mechanisms. First, we examine how metabolic imbalances in the tumor microenvironment mediate immune escape by suppressing immune cell function. The review subsequently analyzes advances in molecular imaging, focusing on the development of novel radiopharmaceuticals that target specific immunometabolism pathways, enabling direct visualization of immune cell dynamics. Furthermore, we analyze advances in novel tracers targeting specific immunometabolism processes for direct immune cell visualization, along with the application of radiomics for early response prediction. Finally, we address current technical and clinical challenges, outlining future directions aimed at constructing an integrated tumor metabolic-immune map and developing robust, PET-based patient stratification strategies to guide personalized therapeutic regimens. In conclusion, PET/CT is emerging as an indispensable multi-dimensional tool, providing critical real-time insights into immunometabolism activity to optimize therapeutic decision-making and improve clinical outcomes in cancer immunotherapy. Integrating these metabolic-immune signatures with other biomarker data holds promise for refining response assessment strategy and advancing personalized oncology.
The comparative effects of non-insulin glucose-lowering therapies on coronary plaque progression in type 2 diabetes (T2DM) remain unclear. This study aimed to evaluate the impact of five major classes of non-insulin therapies on the progression of coronary atherosclerosis using serial coronary computed tomography angiography (CCTA). This was a retrospective, registry-based cohort study analyzing 880 serial CCTA scans from patients with T2DM enrolled in the TOCCATA (Tomography of Coronary Artery Plaque and Treatment) registry. Patients were stratified based on their prescribed therapy: metformin (n = 357), dipeptidyl peptidase-4 (DPP-4) inhibitors (n = 97), glucagon-like peptide-1 receptor agonists (GLP-1 RAs, n = 88), sodium-glucose cotransporter-2 inhibitors (SGLT2i, n = 258), and thiazolidinediones (TZDs, n = 80). The primary endpoints were vessel-level stenosis progression and changes in patient-level plaque scores, including the Maximum coronary stenosis score (MAXS), segment involvement score (SIS), and segment stenosis score (SSS). Multivariable Cox regression models, adjusted for relevant covariates, were used for the analysis. GLP-1 RAs were associated with the most significant reduction in vessel-level stenosis progression (adjusted Hazard Ratio [HR] 0.68, 95
To compare qualitative reader-based image quality and anatomical detail of hepato-bilio-pancreatic structures between photon-counting detector CT (PCCT) and third-generation dual-source energy-integrating detector CT (EID-CT) in the same patient cohort. This retrospective, single-center qualitative study included 25 patients who underwent contrast-enhanced abdominal CT with both PCCT (Siemens NAEOTOM Alpha pro) and EID-CT (Siemens SOMATOM Force) within a 3-month interval using comparable acquisition protocols, with the exception of reconstructed section thickness (0.4–0.6 mm for PCCT vs 1.0 mm for EID-CT). Four independent radiologists (two with > 10 years and two with < 5 years of experience) evaluated six parameters using a five-point Likert scale: overall image quality, arterial vascular visualization, venous vascular visualization, pancreatic ductal tree, biliary ductal tree, and peripancreatic lymph nodes. All readers were blinded to scanner type. Paired t-tests and non-parametric Wilcoxon signed-rank tests compared mean scores between PCCT and EID-CT. Inter-reader agreement was assessed using intraclass correlation coefficients. Statistical significance was set at p < 0.05. No quantitative image metrics or diagnostic performance outcomes were assessed; analyses were limited to reader-based Likert scale image quality scores. PCCT demonstrated statistically significant superiority across all six parameters compared with EID-CT (overall image quality 4.82 vs 3.80, arterial vascular visualization 4.87 vs 3.98, venous vascular visualization 4.84 vs 3.91, pancreatic ductal tree 4.83 vs 3.51, biliary ductal tree 4.87 vs 3.63, peripancreatic lymph nodes 4.86 vs 3.81; all p < 0.0001). Mean score improvements with PCCT ranged from 0.89 to 1.32 points across all parameters. Inter-reader agreement was good to excellent for PCCT (ICC 0.70–0.92) and poor to moderate for EID-CT (ICC 0.43–0.70). Reader experience level did not significantly influence assessments (p = 0.94). These findings derive from a qualitative head-to-head comparison of image quality scores and were consistent across all four readers. In this preliminary intra-patient comparison, photon-counting CT provided significantly superior image quality and anatomical detail in the hepato-bilio-pancreatic region compared to third-generation dual-source EID-CT. Enhanced visualization of ductal structures, distal vessels, and lymph nodes may support improved diagnostic confidence and more precise staging, although diagnostic performance was not assessed in this study. Because reconstructed section thickness differed between systems, these qualitative advantages cannot be attributed to detector technology alone and require confirmation in larger studies with matched protocols and diagnostic-accuracy endpoints.
To evaluate the diagnostic performance of contrast-enhanced CT and CT-based radiomics in detecting thyroid cartilage invasion in patients with laryngeal squamous cell carcinoma (LSCC), using histopathology as the reference standard. This retrospective single center study included 60 patients with histologically confirmed glottic LSCC who underwent pre-treatment contrast-enhanced CT between January 2015 and December 2025. Two head and neck radiologists independently assessed thyroid cartilage invasion based on established imaging criteria. Texture analysis was performed on manually segmented thyroid cartilage volumes, extracting 107 textural features. Feature selection and model construction were performed using LASSO logistic regression, whereas diagnostic performance was evaluated using receiver operating characteristic (ROC) curve. CT-based radiologist assessment showed moderate diagnostic performance in detecting thyroid cartilage invasion, with 63.3
PURPOSE:To compare the diagnostic performance of three different abbreviated MRI protocols simulated by extraction from a standard liver MRI protocol with hepatobiliary-specific contrast agent obtained for hepatocellular carcinoma (HCC) surveillance in patients candidates for liver transplantation. METHODS:We conducted a retrospective study on cirrhotic patient population, undergoing MRI surveillance for HCC. All patients were candidates to liver transplantation with a total of 105/195 patients previously treated with liver-directed therapy. From the complete liver MRI with hepatobiliary-specific contrast agent (cMRI), three simulated abbreviated MRI protocols were extracted: dynamic contrast-enhanced abbreviated MRI (Dyn-AMRI); hepatobiliary phase abbreviated MRI (HBP-AMRI) and non-enhanced abbreviated MRI (NE-AMRI). Three abdominal radiologists independently interpreted the simulated protocols. RESULTS:195 cirrhotic patients, candidates for liver transplantation, were analyzed. The overall sensitivity, specificity and accuracy of AMRI for HCC detection were 94%, 96% and 94% for Dyn-AMRI, 93%, 94% and 93% for HBP-AMRI, 90%, 90% and 90% for NE-AMRI. The interchangeability analysis revealed no significant difference between HBP-AMRI and cMRI and between cMRI and Dyn-AMRI. Regarding equivalence, agreement and interchangeability analysis, no significant differences were found in detection of HCC in patients previously undergoing liver-directed therapy. CONCLUSION:Both Dyn-AMRI and HBP-AMRI could be used in surveillance of cirrhotic patients, even after previous liver-direct therapy.
Artificial intelligence (AI) is increasingly proposed as a solution to improve efficiency in radiology and nuclear medicine, particularly in the context of workforce shortages. However, adoption of AI-based clinical decision support systems (AI-CDSS) remains slow, due to limited model transparency. Explainable AI (XAI) may improve clinician acceptance by supporting oversight and trust. This study evaluated the impact of different XAI explanation types on radiologists’ willingness to adopt AI systems. Ten nuclear medicine radiologists from eight UK institutions performed lung cancer TNM staging using whole-body PET/CT scans supported by a simulated AI-CDSS. Three explanation approaches were assessed: input feature attribution, high-level concept explanations and global algorithmic transparency. Adoption likelihood and explanation usefulness were rated using Likert scales and analysed with nonparametric sign tests. Semi-structured interviews were additionally analysed through thematic evaluation supported by large language model-assisted coding with human verification. All explanation approaches significantly increased radiologists’ willingness to adopt the AI system compared to a black-box model (p < 0.05). Explanations were consistently considered useful in enabling participants to confirm or challenge AI staging recommendations (p < 0.001). Qualitative findings highlighted the importance of clinical relevance, error detection and decision support value. A trade-off between explanation depth and usability was identified as a key factor influencing preferences. Incorporating XAI into nuclear medicine CDSS enhances radiologists’ acceptance and provides clinically meaningful information for oversight of AI recommendations, in accordance with the EU AI Act. These findings support the role of XAI in facilitating integration of AI tools into diagnostic workflows.
Radiology is rapidly evolving from a service that produces images into a data-centric clinical platform that supports prevention, early diagnosis, and personalized care. This shift is accelerated by the convergence of digital health, artificial intelligence (AI), and quantitative imaging approaches such as radiomics. However, the clinical impact of these innovations depends less on algorithms alone and more on robust digital infrastructures and true interoperability across radiological, clinical, and-when available-molecular data. In practice, many implementations fail because data remain fragmented across heterogeneous systems, metadata are incomplete, workflows are not harmonized, and governance frameworks are insufficient to ensure quality, privacy, and accountability. This paper focuses on the radiology-centric requirements for interoperable data ecosystems, covering technical and semantic standards (e.g., DICOM, HL7 FHIR, IHE profiles, controlled terminologies), data governance and quality programs, and AI integration into real-world radiology workflows. Common practical barriers, including legacy IT debt, semantic inconsistencies, model drift, bias, medico-legal uncertainty, and hidden operational costs, are discussed. Last, pragmatic recommendations to enable scalable, secure, and clinically useful interoperability are proposed-supporting prevention pathways and the responsible deployment of AI and radiomics at institutional and network levels.
PURPOSE:Clinical application of late iodine enhancement (LIE) for myocardial scar identification is limited by low contrast-to-noise ratio (CNR). Photon-counting detector-CT (PCD-CT) can improve image quality providing spectral information and low image noise. The aim of the study was to assess the impact of spectral analysis and reconstruction parameters on LIE image quality using PCD-CT. MATERIAL AND METHODS:This single-center retrospective study included 74 patients undergoing PCD-CT between May 2024 and July 2025 for myocardial scar evaluation. Images from patients with visible LIE were reconstructed using four parameter combinations (Qr40 and Qr36 kernels; 0.4-mm and 2-mm slice thickness) and analyzed at monoenergetic levels (40-120 keV, 10 keV increment) using a dedicated workstation. CNR and signal-to-noise ratio (SNR) were calculated for each reconstruction. RESULTS:Among 74 patients, 41 (55%) had myocardial scar (M:F 32:9; 62 years [IQR,56-71]) and 24/41 (59%) had an implantable cardioverter defibrillator (ICD). Scars were predominantly transmural (21/41, 51%). A 40-keV VMI achieved the highest CNR and SNR, regardless of reconstruction parameters (p < 0.05), with median CNR of 5.36 [IQR, 4.07-6.88] using 2-mm slice thickness and Qr36 kernel. Scar pattern, ICD, and BMI did not significantly affect either CNR or SNR (p > 0.05). In the exploratory subgroup of 16 patients undergoing LGE-MRI, CNR is not statistically different between LIE-CT at 40 keV with 2-mm slice thickness from LGE-MRI (5.36 [IQR,4.07-6.88] vs. 8.04 [IQR,6.67-9.89]; p = 0.178). CONCLUSION:PCD-CT LIE demonstrated the highest CNR at 40-keV VMI with 2-mm slice thickness, regardless of scar pattern, BMI, or ICD presence.
To develop and validate a hybrid model integrating intratumoral (TR) and peritumoral (PTR) CT radiomics with clinical and hematologic inflammatory markers for predicting lymph node metastasis (LNM) in esophageal squamous cell carcinoma (ESCC). This retrospective single-center cohort study included 304 patients with pathologically confirmed ESCC. Patients were randomly divided into a training cohort (n = 243) and a test cohort (n = 61) in an 8:2 ratio. Radiomics features were extracted from manually segmented three-dimensional volumes of interest (VOIs) on venous-phase thin-slice contrast-enhanced CT images. The PTR VOIs were defined as 1-, 2-, and 3-mm ring-shaped regions surrounding the TR VOI. Clinical data, hematologic inflammatory markers, and TR and PTR CT radiomics features were used to construct six categories of models: a clinical model, a hematologic model, TR + PTR radiomics models, a clinical + TR + PTR radiomics model, a hematologic + TR + PTR radiomics model, and a clinical + hematologic + TR + PTR hybrid model. Model performance was evaluated using receiver operating characteristic (ROC) analysis, calibration curves, and decision curve analysis (DCA). The overall LNM prevalence was 37.5
PURPOSE:To develop a deep learning-based framework for automated detection and grading of breast arterial calcification (BAC) on mammograms, and to evaluate its association with major adverse cardiovascular events (MACE). MATERIAL AND METHODS:This retrospective case-control study used two datasets: SegModel (1270 mammograms with BAC annotations) and MACEPred (3190 mammographic cases of women with MACE, plus 6458 controls). A U-Net segmentation model and Cox proportional hazards models were used to assess adjusted hazard ratios (HRs) for four BAC grading strategies: binary categorization, area-based grading (none, mild, moderate, severe), intensity-based grading, and a combined approach. FINDINGS:The U-Net attained a Jaccard similarity coefficient of 0.582, accuracy of 0.996, precision of 0.801, F1 score of 0.756, and recall of 0.716 in segmenting BAC. The presence of BAC was associated with an adjusted HR of 1.16 (95% CI 1.10-1.23). For area-based grading, the HRs for mild, moderate, and severe grades were 1.13 (95% CI 1.06-1.20), 1.30 (95% CI 1.18-1.45), and 1.58 (95% CI 1.15-2.18), respectively. Intensity-based grading showed respective HRs of 1.08 (95% CI 1.00-1.17), 1.10 (95% CI 1.01-1.20), and 1.18 (95% CI 1.09-1.28). The combined approach demonstrated respective HRs of 1.102 (95% CI 1.04-1.17), 1.249 (95% CI 1.14-1.37), and 1.654 (95% CI 1.30-2.10) for mild, moderate, and severe grades. CONCLUSION:Our study presents a novel automated framework for BAC assessment that provides independent insights into cardiovascular risk in women. Combined area and intensity grading reflected increasing MACE risk across BAC severity levels, although its incremental predictive improvement was limited.
To compare biparametric MRI (bpMRI) and multiparametric MRI (mpMRI) for detecting clinically significant prostate cancer (csPCa), and to assess the impact of artificial intelligence (AI)–assisted bpMRI on diagnostic performance and biopsy-related outcomes in readers with different expertise. In this retrospective multi-reader study, 173 men referred for prostate mpMRI were evaluated by five radiologists (two experts, three basic readers) who scored bpMRI and mpMRI using PI-RADS v2.1. After a 45-day wash-out, the same readers reinterpreted bpMRI with concurrent AI decision support (available for 127 cases). Diagnostic performance for csPCa (ISUP ≥ 2) and benefit-to-harm ratios were compared across protocols and reader groups. Mean area under the receiver operating characteristic curve was 0.820 for bpMRI and 0.819 for mpMRI (difference 0.001), indicating comparable diagnostic performance. In the AI subset, AI increased mean specificity from 59.7
PURPOSE:To evaluate whether routinely assessed MRI features can reliably estimate the timing of rotator cuff tears (RCTs) and to assess the performance of an MRI-based logistic regression model in distinguishing recent from chronic tears. MATERIAL AND METHODS:In this retrospective single-center study, 255 patients with clinically and MRI-confirmed RCT following shoulder trauma underwent MRI between 2011 and 2024. Tears were classified as acute (< 6 weeks), subacute (6-12 weeks), or chronic (> 12 weeks) based on the reported date of trauma; acute and subacute tears were grouped as "recent". Ten predefined MRI features were independently assessed. Univariable diagnostic performance metrics were calculated for each feature. A multivariable Firth-penalized logistic regression model was developed to discriminate recent from chronic tears, with internal validation performed using bootstrap resampling. RESULTS:Among the 255 patients (mean age 58.3 years; 65% male), intra-/peri-muscular edema (42% in recent vs 12% in chronic tears) and frayed, hyperintense tendon fibers emerged as independent predictors of recent injury (odds ratios 4.15 and 6.78, respectively). However, traditional markers of chronicity-including fatty infiltration, muscle atrophy, tendon retraction, and superior humeral head migration-showed limited discriminative value. The multivariable model demonstrated modest performance, with an ideal area under the curve (AUC) of 0.74 and an optimism-corrected AUC of 0.68. Sensitivity and specificity at the optimal threshold were 56% and 82%, respectively. CONCLUSION:Although selected MRI findings are associated with recent RCT, an MRI-based logistic regression model provides limited accuracy for timing estimation.
To evaluate the efficacy and safety of single-fraction LINAC-based radiosurgery (RS) for patients with drug-resistant trigeminal neuralgia (TN). We retrospectively reviewed 46 patients treated for TN between August 2008 and December 2024, 42 were evaluable. RS was delivered with a linear accelerator equipped with a micro-multileaf collimator. Diagnostic cisternography sequence MRI was co-registered with the CT plan to delineate the target volume and develop the treatment plan. The CTV was contoured on the retro-Gasserian ganglion. The dose was prescribed to the isocentre for all patients. The prognostic impact of parameters such as sex, side of TN, previous surgery, radiotherapy dose, clinical response probability, and response onset time was assessed. Statistical analysis was performed using the MedCalc software package and the Kaplan–Meier product limit method. Acute and late toxicities were graded according to the CTCAE v5.0 scale. Patient characteristics For the 46 treated patients were as follows: 29 out of 46 females (63
Pancreatic cystic lesions represent an increasingly common clinical finding due to the widespread use of cross sectional imaging, encompassing a heterogeneous spectrum ranging from benign entities to premalignant and malignant neoplasms. Their rising incidence poses significant challenges in diagnosis, risk stratification, and management, requiring a careful balance between avoiding unnecessary interventions and preventing malignant progression. This position paper, developed by an intersocietal multidisciplinary panel of experts in pancreatic disease, aims to provide a comprehensive and evidence-based overview of these lesions. Current classifications, imaging features, and differential diagnosis are discussed, with particular emphasis on the role of magnetic resonance imaging. Key entities including serous cystic neoplasms, mucinous cystic neoplasms, intraductal papillary mucinous neoplasms, and rarer cystic tumors are analyzed in terms of biological behavior and radiological characteristics. Furthermore, the document reviews available international guidelines and proposes a pragmatic approach to clinical management, including indications for surveillance, endoscopic evaluation, and surgical treatment. Special attention is given to risk stratification based on high-risk stigmata and worrisome features, as well as to individualized patient management. In conclusion, this paper provides a shared expert perspective to support standardized and personalized management of pancreatic cystic lesions in clinical practice.
Branch-duct intraductal papillary mucinous neoplasms (BD-IPMNs) are pancreatic cystic lesions originating from the pancreatic ducts, characterized by mucin production and progressive ductal dilation. They exhibit a wide spectrum of biological behavior, ranging from indolent lesions to entities with significant malignant potential. Although the 2024 Kyoto guidelines define worrisome features (WF) and high-risk stigmata (HRS) to support risk stratification and clinical management, predicting disease progression remains challenging. In this retrospective study, we investigated whether MRI-based radiomic analysis could identify, at the time of initial imaging, patients with BD-IPMNs who subsequently develop WF or HRS according to 2024 Kyoto guidelines. A total of 194 adult patients who underwent at least two MRI examinations between January 2011 and March 2025 were included, with a median follow-up of 53 months; progression was observed in 28.3
