Journal of Computer Assisted Tomography最新文献

Neuroendocrine malignancies of the cervix (NEM) are an aggressive tumor with early metastasis and recurrence, resulting in poor prognosis and 5-year mortality ranging from 10% to 25% for later stages of disease. These are a rare malignancy of middle-aged and elderly women with an incidence of <2%. The clinical presentation is like the common cervical cancer-squamous cell carcinoma, presenting with nonspecific vaginal bleeding. Pap smear has very low sensitivity and accuracy for NEM, leading to delayed diagnosis. Due to its low incidence, the lack of specific clinical findings, and the limited sensitivity of Pap smears for diagnosing NEM, delays in diagnosis are common, contributing to a higher number of late-stage presentations. Hence, diagnosis by imaging using features such as intratumoral necrosis and hemorrhage, lower ADC values with a cutoff of 0.9×10-3 mm2/s, lymphadenopathy, parametrial and vascular invasion, infiltrative features, and pathologic sampling with immunohistochemical staining is crucial and of paramount importance. We aim to discuss the clinical, imaging, and pathologic findings of NEM and describe the current management options for patients with this aggressive malignancy.

Objective: To quantitatively evaluate MRI-derived morphologic and bursal parameters in patients with calcific tendinopathy (CaT) of the shoulder and to determine imaging thresholds predictive of rotator cuff (RC) tears.

Materials and methods: In this retrospective, single-center study, 112 patients with calcific tendinopathy who met the inclusion criteria were included (January 2020 to April 2025). Patients with CaT confirmed by radiography or computed tomography were included. Exclusion criteria were prior shoulder surgery, incomplete MRI sequences, severe motion artifacts, or calcifications outside the RC tendons. MRI features recorded included tendon involvement, calcification morphology (solitary or multifocal), maximum deposit size, bursal thickness, and presence of bursitis. RC tear presence and severity were assessed in 3 planes on fluid-sensitive sequences. Receiver operating characteristic (ROC) analysis determined optimal thresholds for tear prediction.

Results: Of 134 eligible patients, 112 met inclusion criteria (84 females, 28 males; mean age: 50.3±10.5 y). RC tears were identified in 42.9% of cases. Multifocal morphology was more frequent in patients with tears (79.2% vs. 25.0%, P<0.001). Mean deposit size was greater in the tear group (10.2±3.5 vs. 8.5±2.6 mm, P=0.004); ROC analysis identified a ≥9 mm cutoff (AUC=0.62). Bursal thickness was significantly higher in the tear group (5.1±2.1 vs. 2.8±1.9 mm, P<0.001), with a ≥5 mm threshold demonstrating good predictive value (AUC=0.81). Bursitis was more prevalent in tear cases (97.9% vs. 62.5%, P<0.001).

Conclusion: MRI-derived multifocal calcification morphology, deposit size ≥9 mm, and bursal thickness ≥5 mm are significant predictors of RC tears in CaT. Quantitative assessment of these parameters can enhance MRI interpretation and may guide early management decisions.

Objective: To investigate the potential of the radiomic features derived from the diffusion magnetic resonance imaging (MRI)-based elastography in identifying patients with different liver fibrosis stages.

Methods: Data from 151 patients with liver fibrosis, collected between January 2018 and August 2021, were included. Liver fibrosis staging was determined using histopathology from surgical resection specimens, assessed by the METAVIR scoring system, with significant fibrosis defined as ≥F2 and advanced fibrosis as ≥F3. Shifted apparent diffusion coefficient (sADC) and virtual MR elastography (vMRE) were calculated from the diffusion-weighted imaging (DWI) technique with b values of 200 and 1500 s/mm2. A total of 3222 radiomic features were extracted. A support vector machine was applied to build a classification model with radiomic features. We compared the results of models constructed by the mean values of MRE, radiomic features of sADC, vMRE, and sADC + vMRE in identifying significant and advanced liver fibrosis. The corresponding radiomics scores were also applied in each subject for comparison.

Results: sADC and vMRE data without clinical data provided better classification performance than that of combined features of imaging and clinical data. As compared with MRE, the radiomics model derived from sADC + vMRE yielded higher AUCs of 0.90 and 0.91 in identifying significant and advanced liver fibrosis stages, respectively. Radiomics scores achieved a comparable AUC of 0.91. The radiomics scores of sADC + vMRE were significantly correlated with the pathologic fibrosis stage.

Conclusions: The application of radiomics in sADC and vMRE performed well to identify different stages of liver fibrosis. Radiomics applied to sADC and vMRE may provide complementary information that is associated with histopathologic changes.

Objectives: We hypothesized that interpretive error location and patient harm are significantly associated in a quality assurance database.

Methods: We reviewed all interpretive errors reported in the abdomen and pelvis in a quality assurance database over a 5-year period. Imaging exam data, interpretive error, anatomic location, and clinical outcome by the Agency for Healthcare Research and Quality Harm Scale were recorded. Multivariable logistic regression modeling was performed.

Results: 263 interpretive error reports were identified. The most reported error locations included hepatobiliary (15%), gastrointestinal tract (13%), bone (13%), reproductive organs (12%), and peritoneum/mesentery/omentum (12%). Of the 263, 165 (63%) errors were found to result in any patient harm, compared with 98/263 (37%) errors with no harm. Anatomic category was significantly associated with harm when adjusted for age, sex, exam setting, and exam priority in our logistic regression model (P=0.003). When compared with hepatobiliary errors, errors in the kidney/ureter/bladder, gastrointestinal tract, and reproductive organs were associated with greater odds of harm. Errors in the peritoneum/mesentery/omentum and bone were associated with smaller odds of harm. Retroperitoneum/body wall and vascular errors had similar odds of harm to hepatobiliary errors.

Conclusion: Interpretive error location is significantly associated with patient harm. Our results may inform an evidence-based approach toward improving the process by which errors are reviewed in radiology quality assurance programs. The most reported interpretive error locations in an abdominal imaging quality assurance system are not necessarily the most harmful to patients.

Objective: To assess whether few-shot prompting improves the performance of 2 popular large language models (LLMs) (ChatGPT o1 and DeepSeek-R1) in assigning Coronary Artery Disease Reporting and Data System (CAD-RADS™ 2.0) categories.

Methods: A detailed few-shot prompt based on CAD-RADS™ 2.0 framework was developed using 20 reports from the MIMIC-IV database. Subsequently, 100 modified reports from the same database were categorized using zero-shot and few-shot prompts through the models' user interface. Model accuracy was evaluated by comparing assignments to a reference radiologist's classifications, including stenosis categories and modifiers. To assess reproducibility, 50 reports were reclassified using the same few-shot prompt. McNemar tests and Cohen kappa were used for statistical analysis.

Results: Using zero-shot prompting, accuracy was low for both models (ChatGPT: 14%, DeepSeek: 8%), with correct assignments occurring almost exclusively in CAD-RADS 0 cases. Hallucinations occurred frequently (ChatGPT: 19%, DeepSeek: 54%). Few-shot prompting significantly improved accuracy to 98% for ChatGPT and 93% for DeepSeek (both P <0.001) and eliminated hallucinations. Kappa values for agreement between model-generated and radiologist-assigned classifications were 0.979 (0.950, 1.000) ( P <0.001) for ChatGPT and 0.916 (0.859, 0.973) ( P <0.001) for DeepSeek, indicating almost perfect agreement for both models without a significant difference between the models ( P =0.180). Reproducibility analysis yielded kappa values of 0.957 (0.900, 1.000) ( P <0.001) for ChatGPT and 0.873 [0.779, 0.967] ( P <0.001) for DeepSeek, indicating almost perfect and strong agreement between repeated assignments, respectively, with no significant difference between the models ( P =0.125).

Conclusion: Few-shot prompting substantially enhances LLMs' accuracy in assigning CAD-RADS™ 2.0 categories, suggesting potential for clinical application and facilitating system adoption.

Objective: Whole-brain computed tomography perfusion (CTP) imaging is a dose-intensive imaging technique. We aimed to investigate optimal scanning protocol of the whole-brain CTP using a 320-detector row CT in reducing radiation dose for acute ischemic stroke (AIS) patients.

Methods: This study included 54 consecutive AIS patients who underwent whole-brain CTP on a 320-detector row CT scanner. We evaluated the penumbra and ischemic core volumes of CTP full , CTP 3/4 , and CTP 1/2 , created using full, 3/4 and 1/2 scanning data, respectively. Wilcoxon signed-rank test, Spearman correlation coefficient, and Bland-Altman analysis were used for the statistical analysis. In addition, hypothetical treatment decisions based on the DEFUSE-3 criteria were also evaluated to determine whether there were any differences in the treatment decisions when using reduced sampling data (CTP 3/4 and CTP 1/2 ) compared with full data to assess its clinical efficacy.

Results: The penumbra and ischemic core median volumes on CTP full , CTP 3/4 , and CTP 1/2 were 111.5 mL [interquartile range (IQR): 52.0-173.0] and 5.5 mL (IQR: 0 to 24.0), 106.5 mL (IQR: 47.0 to 170.0) and 6.5 mL (IQR: 0 to 24.0), 106.5 mL (IQR: 48.0 to 178.0), and 5.5 mL (IQR: 0 to 23.0), respectively. There were no significant differences in penumbra ( P >0.05) and ischemic core ( P >0.05) volumes between CTP full , CTP 3/4 , and CTP 1/2 . Spearman correlation analysis showed significant correlations between CTP full and CTP 3/4 and CTP 1/2 for both penumbra ( r =0.989 to 0.998, P <0.001) and ischemic core ( r =0.997 to 0.982, P <0.001) volumes. The hypothetical treatment strategies determined using reduced sampling data (CTP 3/4 , and CTP 1/2 ) were largely consistent compared with those using CTP full .

Conclusions: The use of half-scanning data for the whole-brain CTP image with a 320-detector row CT may help to lower the radiation exposure to AIS patients without significant loss of perfusion information.

Purpose: To characterize the predictive value of CT findings of fluid overload for predicting survival in patients undergoing transcatheter aortic valve replacement (TAVR).

Materials and methods: A retrospective review was performed on 265 patients undergoing CTA for TAVR planning purposes. Images for each patient were analyzed for evidence of fluid overload. Additional clinical data were gathered for each patient including serum NT-proBNP, eGFR, and albumin along with echocardiographic evaluation of left ventricular systolic function. Survival between groups with and without CT evidence of fluid overload (CTFO) was compared using Kaplan-Meier survival analysis and Cox proportional hazards model.

Results: Kaplan-Meier analysis demonstrates survival differences between the subjects with and without evidence of fluid overload. The Cox model demonstrates that CTFO is an independent predictor of survival outcomes. The hazard ratio in a model accounting for multiple variables was 2.93 with a P -value of 0.01. Notably, the Kaplan-Meier analysis demonstrates 100% survival for the first 50 days in patients with euvolemia on CT.

Conclusions: CT evidence of fluid overload before TAVR is associated with increased mortality.