Objective: To aim of this study was to compare the performance of different iterative metal artifact reduction (iMAR) algorithms at reducing metal artifacts after lumbar spine internal fixation surgery. Methods: Patients who underwent internal lumbar spine fixation were prospectively enrolled. Image reconstruction was performed without the iMAR algorithm and with eight recommended iMAR algorithms tailored for different clinical scenarios (neurocoils, dental fillings, spinal implants, shoulder implants, pacemakers, thoracic coils, hip implants, and extremity implants). The regions of interest (ROI), namely, the vertebral body, spinal canal, erector spinae, and subcutaneous fat, were delineated on optimal cross-sectional images along the long axis of the screws, as well as on cross-sectional images without obvious artifacts. The computed tomography (CT) values and their standard deviation (SD) were recorded, and the CT value difference (ΔCT) of each region, average background SD, average artifact SD, and artifact index (AI) value of each region were calculated. Tanking images without the iMAR algorithm were used as the control group, and those reconstructed with the iMAR algorithm were used as the experimental group. The effectiveness of metal artifact removal and its impact on diagnosis were assessed subjectively and objectively. Subjective and objective evaluation results were compared. Results: Thirty-six patients, including 22 males (61.11%), were included in the study, with a mean (age±SD) of (62.75±12.23) years. The median (interquartile range) of the volume CT dose index value (CTDIvol) was 16.25 (14.55, 19.62) mGy, and the median (interquartile range) of the dose-length product (DLP) was 448.45 (410.52, 566.88) mGy·cm. The ΔCT values of each region, average artifact SD value, and AI values of each region were significantly lower in images reconstructed with the spinal iMAR algorithm than those reconstructed without the iMAR algorithm. The muscle ΔCT value, fat ΔCT value, average artifact SD value, and AI values of each region were significantly lower in images reconstructed with the spinal iMAR algorithm than in those reconstructed using other iMAR algorithms. The subjective scores were significantly higher for images reconstructed with the spinal iMAR algorithm than for those reconstructed without the iMAR algorithm. Conclusion: Compared with the iMAR algorithms used for other clinical scenarios, the iMAR algorithm for spinal implants was more effective at removing metal artifacts in patients after lumbar spine internal fixation surgery and is recommended for clinical diagnosis.
Objective: To investigate the clinical value of ultra-high-resolution computed tomography (U-HRCT) in diagnosing extremity joint fractures and to compare its performance with that of multi-slice spiral computed tomography (MSCT). Methods: A prospective cohort study was conducted on 135 patients with suspected extremity joint fractures who presented to Nanjing Tongren Hospital between February 2024 and February 2026. All patients underwent sequential imaging using 16-slice MSCT and domestic U-HRCT systems. The gold standard for surgical results or follow-up CT/MRI at 2–4 weeks was used as a reference. We compared the detection rates, miss rates, incremental diagnostic yields, and image quality scores (assessed using a 5-point Likert scale) between the two modalities using subgroup analysis across different anatomical locations. Results: In total, 245 fractures were identified. MSCT identified 169 fractures, yielding a detection rate of 68.98% (169/245) and miss rate of 31.02%. Notably, U-HRCT detected 241 fractures, achieving a significantly higher detection rate of 98.37% (241/245) and a substantially lower miss rate of 1.63%. U-HRCT also identified 72 fractures missed by MSCT, including 28 intra-articular fractures (38.89%) and 25 avulsion fractures (34.72%), representing an incremental detection rate of 29.39%. The detection rates of U-HRCT for hand, wrist, elbow, ankle, and foot fractures were all significantly higher than those of MSCT. Image quality scores revealed that U-HRCT provided superior delineation of the cortical bone, trabecular microstructure, and articular surfaces compared with MSCT, with no false-positive cases observed. Conclusion: U-HRCT significantly improved the detection of extremity joint fractures, particularly subtle and occult lesions, effectively resolving the diagnostic uncertainties inherent in conventional MSCT. This provides a reliable imaging basis for precise fracture classification and informed surgical decision making.
Objective: To develop a machine learning model integrating MRI visually accessible Rembrandt images (VASARI) features and apparent diffusion coefficient (ADC) parameters for the preoperative classification of the three subtypes of adult-type diffuse glioma. Methods: This study retrospectively enrolled 206 patients with pathologically confirmed adult-type diffuse glioma and divided them by scanner type into a training set (n=114) and an independent validation set (n=92). Tumor multi-region segmentation was conducted using nnU-Net, followed by extraction of VASARI features and ADC-related parameters. Recursive feature elimination with cross-validation (RFECV), XGBoost feature importance ranking, and incremental feature selection were applied to identify the optimal feature subset, and an XGBoost-based multiclass fusion model was subsequently constructed. Results: Twelve VASARI features and four ADC parameters (ADCmin, ADCmean, rADCmin, and rADCmean) differed significantly among the three subtypes. Eight features were finally selected for model construction. The fusion model achieved accuracies of 96.49% in the training set and 80.43% in the validation set. In the validation cohort, the area under the curves (AUCs) for predicting astrocytoma and glioblastoma were 0.96 and 0.97, respectively, both higher than those of the standalone VASARI and ADC models, while the AUC for oligodendroglioma was 0.59. SHAP analysis revealed that diffusion restriction, ADC parameters, and the margin of the non-enhancing region were the primary contributors to the model. Conclusion: The XGBoost-based fusion model integrating MRI-VASARI features and ADC demonstrated a high diagnostic performance in distinguishing glioblastoma and astrocytoma, but its ability to differentiate oligodendroglioma remained limited. Future studies should incorporate additional imaging features, such as calcification, to further improve the model performance.
Objective: To investigate the impact of the orthopedic metal artifact reduction (O-MAR) technique on the measurement of quantitative parameters across various tissues in spectral CT images and define the intrinsic algorithmic bias in a metal-free background. Methods: A CIRS 062M electron density phantom was scanned repeatedly (n=10) using a Philips IQon Spectral CT. Quantitative results for eight tissue-simulating inserts (inspiratory lung, expiratory lung, adipose, breast, muscle, liver, trabecular bone, and dense bone) were analyzed with and without the O-MAR algorithm. Reconstructed images included conventional polyenergetic images (Con), calcium-suppressed (CaSupp) images (with suppression indices of 25, 50, and 75), effective atomic number (Zeff) maps, and electron density (ED) maps. Measurements without O-MAR were used as the reference, and Bland–Altman analysis was performed to quantify the mean bias and 95% limits of agreement (95% LoA), supplemented by paired t-tests to evaluate statistical differences. Results: The impact of O-MAR on quantitative measurements exhibited marked tissue specificity: (1) Consistency Evaluation: Inspiratory lung, adipose, and breast tissues demonstrated high consistency with minimal bias. In contrast, the dense bone tissue showed significant fluctuations, particularly in CaSupp 75 images, where the mean bias reached 2.05 HU and 95% LoA was substantially wider than that of other tissues. (2) Difference Testing: Measurements for inspiratory lung, muscle, and adipose tissues showed significant differences across most image types. The dense bone tissue was significantly affected in Con and CaSupp 50/75 images. Conversely, Zeff and ED maps exhibited the highest numerical stability. Conclusion: In a metal-free background, the O-MAR algorithm introduced tissue-specific systematic quantitative biases, with significant impacts on extremely low-density and high-density tissues. However, quantitative parameters in the Zeff and ED maps remained stable.
Objective: To investigate the performance of an artificial-intelligence–based metal artifact correction algorithm (AIMAC) on pelvic computed tomography (CT) after total hip arthroplasty. Methods: This retrospective study included data from 34 patients with a history of hip arthroplasty who underwent pelvic CT at our institution between September 2022 and December 2024. Raw data were reconstructed using three methods: routine reconstruction (Karl 3D, level 5), conventional metal artifact correction (MAC), and AIMAC. For subjective evaluation, two radiologists independently scored overall image quality using a five-point scale in a blinded manner, and inter-reader agreement was assessed using the Kappa statistic. For objective evaluation, regions of interest were drawn to calculate CT-number differences between the artifact-affected and reference slices (ΔCTmuscle and ΔCTfat), standard deviation (SDartifact and SDreference), the metal artifact index (MAI), and the contrast-to-noise ratio (CNR). Overall comparisons among the three reconstructions were performed using repeated-measures analysis of variance or the Friedman test, according to normality. Pairwise comparisons were conducted using the paired t-test or the Wilcoxon signed-rank test, with Holm–Bonferroni adjustment for multiple comparisons (P*). Results: Inter-reader agreement for overall image quality was good (Kappa: 0.866, 0.816, and 0.794). Overall subjective scores differed significantly among the three methods, with AIMAC scoring higher than MAC, and MAC scoring higher than routine reconstruction. For objective metrics, the ΔCTmuscle, ΔCTfat, SDartifact, MAI, and CNR all had significant differences among the three methods, with AIMAC outperforming MAC and routine reconstruction. For SDreference, MAC was lower than routine reconstruction and AIMAC, whereas no significant difference was observed between routine reconstruction and AIMAC. Conclusion: For pelvic CT after hip arthroplasty, AIMAC more effectively reduced metal artifacts, improved CT-number stability, and enhanced overall image quality compared with routine reconstruction and conventional MAC.
