Background: Spread through air spaces (STAS) is associated with recurrence after limited resection of lung adenocarcinoma; however, reliable preoperative assessment remains challenging. Objectives: This study investigated whether routine computed tomography (CT) morphology and tumor CT-derived extracellular volume fraction (CT-ECV) contribute to the preoperative identification of STAS in 1 - 3 cm CT-solid lung adenocarcinomas. Patients and Methods: This retrospective study included 104 surgically resected tumors, of which 53 (51.0%) showed STAS. Twelve prespecified preoperative variables were analyzed using least absolute shrinkage and selection operator (LASSO) logistic regression with repeated nested stratified five-fold cross-validation. A full-cohort stratified 10-fold analysis was performed to describe variable retention at the minimum-error and one-standard-error penalty levels. Out-of-fold discrimination and calibration were assessed using the area under the receiver operating characteristic curve (AUC), Brier score, calibration intercept and slope, and a bootstrap confidence band. Interobserver agreement and CT-ECV remeasurement reproducibility were assessed using Cohen's kappa, the intraclass correlation coefficient (ICC), and Bland-Altman analysis. Results: At the minimum-error penalty, LASSO retained peripheral ill-defined opacity, tumor CT-ECV, and vascular convergence; at the one-standard-error penalty, only peripheral ill-defined opacity was retained. The out-of-fold AUC was 0.632 (95% confidence interval [CI], 0.522 - 0.737), and the mean fold-specific AUC was 0.628 ± 0.122. Calibration was limited (Brier score, 0.237; calibration intercept, 0.016; calibration slope, 0.645). Interobserver agreement for peripheral ill-defined opacity was almost perfect (kappa, 0.919; 95% CI, 0.834 - 0.981). Region-of-interest-based CT-ECV remeasurement on the same CT examinations showed good relative agreement (ICC, 0.890) but wide absolute limits of agreement (−8.62% to 8.91%). Conclusion: In surgically resected 1 - 3 cm CT-solid lung adenocarcinoma, peripheral ill-defined opacity was the most stable contributor to STAS assessment, whereas CT-ECV was less consistent. Modest discrimination and calibration limit clinical utility and support the need for validation in larger, multicenter cohorts before application beyond this population.
Background:Three-dimensional (3D) volumetric magnetic resonance imaging (MRI) is the reference standard for assessing plexiform neurofibromas (PNs) in children with neurofibromatosis type 1 (NF1). However, simplified one-dimensional (1D) and two-dimensional (2D) measurements are often used in clinical practice despite limited validation in pediatric populations. Objectives:This study aimed to assess the reliability of 1D, 2D, and 3D measurements; evaluate their correlation and individual-level agreement; and provide practical time-efficiency data to guide clinical implementation. Methods:This retrospective study analyzed 39 pediatric patients with NF1 and measurable PNs. Lesions were measured using three methods: 1D (longest diameter), 2D (product of the two largest perpendicular diameters), and semi-automated 3D volumetry. Reliability was assessed for each method, and associations among methods were evaluated. Individual-level agreement between 2D estimates and 3D measurements was examined, and the predictive value of 1D and 2D measurements for 3D volume was analyzed using linear regression. The time required for each method was recorded. The primary outcome was the agreement between 2D and 3D measurements because this directly informs clinical substitutability. Results:Intra-observer reliability was excellent for all three methods, with intraclass correlation coefficient (ICC) values of 0.998 for 3D volume, 0.998 for 2D area, and 0.989 for 1D length. Inter-observer reliability was also excellent for 3D volume (ICC = 0.966) and 2D area (ICC = 0.956), and good to excellent for 1D length (ICC = 0.934). Semi-automated 3D volumetry required approximately 4 minutes per lesion. Although 2D area correlated strongly with 3D volume (rho = 0.865, P < 0.001), Bland-Altman analysis of the log-transformed data showed a mean bias of 0.66 on the log scale, with wide 95% limits of agreement (LoA) ranging from -0.78 to 2.09, indicating that strong correlation reflects rank-order consistency rather than individual-level interchangeability. Therefore, 2D-derived estimates cannot reliably substitute for direct 3D measurement at the individual-patient level. Conclusions:Despite the high reliability of all methods, regression analysis on the log-log scale confirmed that simplified 1D and 2D metrics lack sufficient individual-level agreement to replace 3D volumetry for precise PN assessment in children. We therefore recommend 3D volumetric MRI as the reference standard for therapeutic monitoring and trial endpoints. Efficient 2D area measurements can support rapid surveillance and group-level trend analyses, whereas 1D length, despite excellent inter-observer reliability (ICC = 0.934), should be used with caution for individual quantification because of its limited concordance with 3D volumetry in regression analysis. This approach balances precision and feasibility for individualized radiological assessment of NF1-associated PNs in children.
Background: Depth of invasion (DOI) in oral tongue squamous cell carcinoma (OTSCC) is an important prognostic factor influencing tumor staging, surgical planning, and overall prognosis. Objectives: This study aimed to assess the correlation between computed tomography-based radiologic DOI (rDOI) and pathologic DOI (pDOI) and to evaluate the diagnostic accuracy of computed tomography-based rDOI, using pDOI as the gold standard. Patients and Methods: This retrospective, single-center diagnostic accuracy study included 60 consecutive eligible patients with biopsy-proven OTSCC who underwent preoperative contrast-enhanced computed tomography at an academic referral center between April 2022 and March 2023. Patients with recurrent tumors, prior head-and-neck radiotherapy, computed tomography artifacts precluding measurement, or missing pathological DOI were excluded. The index test, computed tomography-derived rDOI, was measured independently by two blinded radiologists with 5 and 20 years of head-and-neck imaging experience and compared with histopathological DOI as the reference standard, assessed by a dedicated head-and-neck pathologist blinded to imaging findings. The primary analysis evaluated agreement and diagnostic accuracy. Inter-reader reliability and the association between DOI and nodal involvement were prespecified secondary objectives. Results: Sixty patients were analyzed, including 35 men and 25 women, with a mean age of 54.0 ± 14.2 years. The mean pDOI was 10.71 ± 6.47 mm. Computed tomography systematically overestimated the pDOI by 1.12 mm for Radiologist 1 and by 2.19 mm for Radiologist 2. Lymph node involvement was present in 15 patients (25%), who had a significantly higher pDOI than those without lymph node involvement (14.13 vs. 8.91 mm; P = 0.002). Strong agreement was observed between pDOI and rDOI, particularly for DOI > 5 mm. For Radiologist 1, κ = 0.797 (95% confidence interval [CI]: 0.69 - 0.91), Spearman's rho = 0.668 (95% CI: 0.53 - 0.77), and P < 0.001. For Radiologist 2, κ = 0.700 (95% CI: 0.59 - 0.81) and P < 0.001. For detecting DOI > 5 mm, computed tomography showed an area under the curve of 0.934 for Radiologist 1 and 0.921 for Radiologist 2, with sensitivity > 89% and specificity > 85%. Inter-reader agreement was excellent (κ = 0.838; 95% CI: 0.76 - 0.92; Spearman's rho = 0.945; P < 0.001) and was independent of patient sex or lymph node status. Conclusion: Computed tomography-based rDOI is a reliable and diagnostically accurate preoperative tool for OTSCC staging, particularly for tumors with DOI > 5 mm. It may serve as a useful preoperative surrogate estimate of DOI when surgical pathology is not yet available. Ultrasound may be more accurate for superficial lesions with DOI ≤ 5 mm.
Context: Artificial intelligence (AI) has a critical role in hospitals that have successfully implemented it and achieved international rankings, because these institutions can serve as benchmarks for other hospitals. The main objective of this article was to examine the characteristics of pioneer hospitals in AI, with a focus on radiology departments. Evidence Acquisition: This narrative review was performed according to SANRA guidelines. Data were collected by searching PubMed, Google Scholar, and ScienceDirect for articles published between January 2019 and June 2025, using search strings that included ('artificial intelligence' OR 'AI') AND ('pioneer hospital') AND ('adoption' OR 'implementation') AND ('radiology'). After screening and selecting 41 papers for the final study, the characteristics of pioneer hospitals were determined. Two independent reviewers then confirmed the common features of pioneer hospitals through discussion and consensus. Risk of bias and reliability were assessed based on study design and transparency of reporting, and grey literature was used cautiously and clearly marked with access dates. Results: Hospital type, innovation, transparency, strong leadership, learning, and hospital size were identified as characteristics of pioneer hospitals. These hospitals have used AI in radiology departments to analyze data, predict outcomes, support clinical decision-making and operational efficiency, and improve diagnostic capabilities and workflow. Conclusion: Successful AI integration in pioneer radiology departments depends on synergy between technical readiness and organizational factors. Hospitals can enhance diagnostic accuracy and operational efficiency by prioritizing strong leadership, a collaborative culture, and a clear digital strategy alongside technological investment.
