Emergency Radiology最新文献

Purpose: This study aimed to develop an automated early warning system using a large language model (LLM) to identify acute to subacute brain infarction from free-text computed tomography (CT) or magnetic resonance imaging (MRI) radiology reports.

Methods: In this retrospective study, 5,573, 1,883, and 834 patients were included in the training (mean age, 67.5 ± 17.2 years; 2,831 males), validation (mean age, 61.5 ± 18.3 years; 994 males), and test (mean age, 66.5 ± 16.1 years; 488 males) datasets. An LLM (Japanese Bidirectional Encoder Representations from Transformers model) was fine-tuned to classify the CT and MRI reports into three groups (group 0, newly identified acute to subacute infarction; group 1, known acute to subacute infarction or old infarction; group 2, without infarction). The training and validation processes were repeated 15 times, and the best-performing model on the validation dataset was selected to further evaluate its performance on the test dataset.

Results: The best fine-tuned model exhibited sensitivities of 0.891, 0.905, and 0.959 for groups 0, 1, and 2, respectively, in the test dataset. The macrosensitivity (the average of sensitivity for all groups) and accuracy were 0.918 and 0.923, respectively. The model's performance in extracting newly identified acute brain infarcts was high, with an area under the receiver operating characteristic curve of 0.979 (95% confidence interval, 0.956-1.000). The average prediction time was 0.115 ± 0.037 s per patient.

Conclusion: A fine-tuned LLM could extract newly identified acute to subacute brain infarcts based on CT or MRI findings with high performance.

Purpose: Hypertensive hemorrhage is the most common type of nontraumatic intracerebral hemorrhage (ICH), and it characteristically originates in deep structures, particularly the basal ganglia, internal capsules, thalami, brainstem, and cerebellum. While advanced imaging modalities like MRI can help uncover culprit lesions in cases of unexplained ICH, we hypothesized that the yield of brain MRI would be low in patients with spontaneous deep intracerebral hemorrhage.

Methods: With IRB approval, we retrospectively reviewed cases of deep ICH at a single tertiary care academic center over a 5-year period and excluded cases with a known cause for hemorrhage. Patient history and demographics, initial blood pressure, and the results of the initial noncontrast head CT and subsequent imaging studies were recorded.

Results: 222 patients met study inclusion criteria, with a median age of 67 and 43.2% female sex. 188 patients (84.7%) had a history of hypertension, while 14 (6.3%) had a urine drug screen positive for cocaine or amphetamines during their hospital admission. The majority of hemorrhages were centered in the basal ganglia or internal capsules (116, 52.3%). Brain MRI was obtained for 120 (54.1%) of cases at a median interval of 0.97 days following the initial head CT, and of these studies, 85 (70.8%) included postcontrast imaging. Only 1 MRI study (0.8%) identified a culprit lesion adjacent to a cerebellar hematoma, which was later found to represent a pilocytic astrocytoma. 33.8% of patients overall met the modified Hong Kong Rule. Of the 77 MRIs performed in patients not meeting the modified Hong Kong Rule, 0 revealed a culprit lesion.

Conclusion: Brain MRI obtained in the acute evaluation of patients with spontaneous deep intracerebral hemorrhage rarely uncovers a culprit lesion. Routine ordering of MRI in this cohort should be reconsidered, particularly in patients not meeting the modified Hong Kong Rule.

Purpose: The use of computed tomography (CT) in the emergency department (ED) has been increasing due to its diagnostic value for emergency physicians (EPs). This study aimed to determine the predictors of EP interpretation errors (IEs) on CT scans leading to change in clinical management (IECM) in both endogenous and exogenous ED visits.

Methods: This single-center, retrospective cohort study included patients with consecutive ED visits initially managed by EPs at our institution over 6 months. Patients who did not undergo CT imaging and presented with cardiopulmonary arrest upon arrival were excluded. CT images were interpreted by emergency radiologists immediately after acquisition, and IEs were identified. The primary outcome was IECM, determined by reference to the clinical management decisions made by EPs. A multivariate analysis was performed to determine the independent predictors of IECM.

Results: Among the 2,037 patients, 158 (8%) had IEs, whereas 52 (3%) had IECM. Multisite CT imaging was the strongest independent predictor for both IECM (OR: 2.25, 95% CI: 1.21-4.19, P = 0.011) and IEs (OR: 2.32, 95% CI: 1.61-3.36, P < 0.001). Other predictors of IECM were prolonged ED stay and night-time ED visits as clinical factors. Additional predictors of overall IEs were contrast-enhanced CT and abdominopelvic CT as radiological factors.

Conclusion: Multisite CT imaging, which involve multiple organs and extensive diagnostic information, significantly increases the likelihood of misinterpretation, leading to change in clinical management by EPs.

Background: Prompt diagnosis of strangulated bowel obstruction (SBO) is critical because delayed recognition can lead to life-threatening complications. This study assessed whether the intestinal-to-liver CT attenuation value ratio-a comparison of ischemic bowel-wall enhancement to liver enhancement-can predict the need for intestinal resection in SBO patients.

Materials and methods: We retrospectively analyzed 52 patients who underwent emergency surgery for suspected SBO from 2014 to 2021. Of these, 35 required intestinal resection due to irreversible ischemia (resection group), while 17 did not (no-resection group). Preoperative clinical and imaging findings were compared between groups.

Results: The resection group had a longer time from onset to surgery (p = 0.034) and higher leukocyte counts (p = 0.037). CT values of the poorly enhanced intestinal wall and the intestinal-to-liver attenuation ratio were significantly lower in the resection group (p < 0.0001). Multivariate analysis identified time to surgery (OR 5.08; 95% CI 1.106-23.350; p = 0.037) and CT attenuation ratio (OR 15.50; 95% CI 2.622-91.686; p = 0.0025) as independent predictors of resection. When stratified by the median ratio cutoff (< 0.40 vs. ≥ 0.40), resection rates were 92% and 44%, respectively (p = 0.0001). Additionally, CT attenuation ratio had the diagnostic performance (AUROC 0.886; Youden index 0.736; sensitivity 97.1% and specificity 76.5%.) CONCLUSION: An intestinal-to-liver CT attenuation ratio below 0.40 is a strong predictor of intestinal ischemia requiring resection in SBO patients.

Purpose: Despite technical advancements in left ventricular assist devices (LVADs), driveline infections (DLIs) remain a common complication evaluated by CT. The purpose of this study was to assess CT imaging features and clinical variables associated with operative versus non-operative management of LVAD DLIs.

Materials/methods: This study analyzed 129 patients with LVAD driveline infections evaluated using CT. Two radiologists assessed CT scans for superficial and deep soft tissue stranding and fluid collections. Logistic regression was used to identify predictors of operative management using imaging and clinical variables, guided by Akaike information criterion. Results were reported as odds ratios, and Interreader agreement was evaluated using Cohen's Kappa.

Results: Operative management was performed in 46.8% of patients. Positive driveline cultures (94.8% vs. 43.5%, p < 0.001) and new antibiotic use (98.3% vs. 72.7%, p < 0.001) were strongly associated with operative intervention. Mild subcutaneous fat stranding was the most frequent CT finding (62.6% and 66.9% by Readers 1 and 2, respectively), whereas deep fluid collections were rare (4.8-5.6%). Clinical predictors of operative management included new antibiotic use (p = 0.036), positive cultures (p < 0.001), and LVAD type. The resulting model achieved an AUC of 0.851 and overall accuracy of 78.6%. The absence of superficial fat stranding on CT significantly predicted non-operative management (p < 0.001).

Conclusion: Positive driveline cultures, recent antibiotic initiation, and absence of skin or subcutaneous fat stranding on CT were associated with non-operative management in LVAD-related driveline infections. Absence of superficial fat stranding on CT may help distinguish suspected driveline infections that are unlikely to require surgical intervention.

Purpose: The study aimed to evaluate the diagnostic accuracy of chest CT for COVID-19 pneumonia in resource-limited Gaza. It compared CT performance to RT-PCR and examined how CT severity scores and interobserver agreement influence diagnostic accuracy, reproducibility, and clinical utility for early detection and triage.

Methods: A retrospective analysis was performed on 252 consecutive patients diagnosed with COVID-19 pneumonia between September 2020 and June 2021 at three major governmental hospitals across the Gaza Strip. Chest CT scans were compared to RT-PCR as the gold standard for diagnosis. CT severity scores were calculated using a 25-point system, and interobserver agreement was assessed using kappa statistics. Sensitivity, specificity, and predictive values were calculated for various threshold levels.

Results: Among the 252 patients included in the study, the mean age was 56.81 ± 11.34 years, with 113 males and 139 females. The diagnostic sensitivity of chest CT was 91.4%, with a specificity of 76.4%. The highest accuracy was observed with a severity score threshold of ≥ 15, with a Youden index of 0.630. Interobserver agreement was excellent (kappa = 0.87) for ground-glass opacities and consolidation. The NPV was 71.2%, indicating the need for supplementary RT-PCR testing in low-prevalence cases.

Conclusion: Chest CT is a reliable diagnostic adjunct for COVID-19 pneumonia, especially in Gaza's severely resource-limited setting, where CT was more accessible than RT-PCR. A CT severity score threshold of ≥ 15 offers an optimal balance of sensitivity and specificity. These findings highlight the practical role of CT imaging in pandemic response in resource-constrained environments.

Background: The triple rule-out computed tomography angiography (TRO-CTA) has recently emerged as a technique that noninvasively evaluates the coronary arteries (CAs), the pulmonary arteries (PAs) and the thoracic aorta (TA).

Objective: To evaluate the feasibility of an optimized scanning protocol to reduce the volume of iodine contrast media (ICM), injection rate, and radiation dose in patients undergoing TRO-CTA.

Methods: Patients undergoing TRO-CTA were assigned to either group A or group B using a 16 cm wide-detector CT. Patients in group A were imaged with a traditional triple scanning protocol with a sequence of the PA, CAs, and TA. Patients in group B were imaged using the modified protocol with scanning sequence of PA, TA, and CAs, ICM of 55 ml, and injection rate of 4.5 mL/s. The image quality and effective radiation dose (ED) were compared.

Results: There were no significant differences in basic information between groups A and B. Other than the left PA, RA, and RV, there were no significant differences in the CT attenuation values of relevant vascular structures between groups A and B. There were no significant differences in CNR values between the two groups except the LAD-D and LCX. The image quality scores were comparable between groups A and B except the CAs. However, there were significant differences between the two groups in ICM (p < 0.05), scanning time (p < 0.001) and ED (p = 0. 023).

Conclusions: The optimized TRO-CTA scanning protocol can achieve less ICM and lower ED while maintaining image quality.

Purpose: Headache is common at emergency services and neuroimaging can help in timely diagnosis of life-threatening pathologies. We evaluated clinical indicators associated with abnormal neuroimaging in patients with acute headache, aiming to develop a scoring system with reliable diagnostic performance.

Methods: This analytical and retrospective study was conducted at a teaching tertiary care hospital in Cali, Colombia, from January 2011 to December 2019. Patients aged 18 years or older with non-traumatic headaches who attended the emergency department and underwent neuroimaging were included. Demographic and clinical data were recorded, including headache associated signs and symptoms, imaging diagnosis and disposition. Statistically significant variables and clinically relevant variables were selected. Data was analyzed using a combination of logistic regression and Receiver Operator Characteristic (ROC) curves, leading to the derivation of three models.

Results: 626 patients were included, 15.5% with abnormal neuroimaging. The variables with the highest odds ratio (OR) were: age > 40 years (OR 3.2 CI 1.86-5.56), motor deficit (OR 5.4 CI 2.62-11.18), visual deficit (OR 3.2 CI 1.56-6.63) and gait disturbance (OR 2.27 CI 0.87-5.96). Three abnormal neuroimaging prediction logistic regression models have been derived. The better scale is performed with model 1, which is validated internally and a cut-off point of 0.179, the Area Under the Curve (AUC) of 0.757 is obtained with a diagnostic accuracy of 0.79 (0.73-0.85).

Conclusion: Our straightforward scale incorporates clinical factors associated with abnormal neuroimaging, with the aim of improving diagnostic performance and predictive capacity to distinguish patients who require neuroimaging.

Acute abdominal complaints constitute up to 40% of all emergency department (ED) presentations in oncology patients due to a multitude of causes. Small bowel pathologies present a diagnostic challenge due to their diverse range and frequently overlapping clinical presentation. In oncology patients, structural changes resulting from tumor growth, surgery and treatment effects can further complicate the diagnostic process. Due to a weakened immune system, oncology patients are also highly susceptible to infections of the gastrointestinal tract (GIT). Traditional computed tomography (CT) scans are used as the gold standard diagnostic modality. However, three-dimensional (3D) postprocessing techniques including maximal intensity projection (MIP), volume rendering (VR) and cinematic rendering (CR) have been employed to aid image evaluation. For a balanced and organized approach to describe diagnostic challenges in this complex population, we have divided the pictorial essay into two parts. The first part focuses on tumor- and infection-associated causes, as summarized below in the visual abstract. The second part will address treatment-related complications, including chemotherapy, radiotherapy, immunotherapy, graft-versus-host disease and post-surgical complications.

Acute ischemic stroke (AIS) is a major cause of mortality and morbidity, with hemorrhagic transformation (HT) as a severe complication. Accurate prediction of HT is essential for optimizing treatment strategies. This review assesses the accuracy and utility of deep learning (DL) and radiomics in predicting HT through imaging, regarding clinical decision-making for AIS patients. A literature search was conducted across five databases (Pubmed, Scopus, Web of Science, Embase, IEEE) up to January 23, 2025. Studies involving DL or radiomics-based ML models for predicting HT in AIS patients were included. Data from training, validation, and clinical-combined models were extracted and analyzed separately. Pooled sensitivity, specificity, and AUC were calculated with a random-effects bivariate model. For the quality assessment of studies, the Methodological Radiomics Score (METRICS) and QUADAS-2 tool were used. 16 studies consisting of 3,083 individual participants were included in the meta-analysis. The pooled AUC for training cohorts was 0.87, sensitivity 0.80, and specificity 0.85. For validation cohorts, AUC was 0.87, sensitivity 0.81, and specificity 0.86. Clinical-combined models showed an AUC of 0.93, sensitivity 0.84, and specificity 0.89. Moderate to severe heterogeneity was noted and addressed. Deep-learning models outperformed radiomics models, while clinical-combined models outperformed deep learning-only and radiomics-only models. The average METRICS score was 62.85%. No publication bias was detected. DL and radiomics models showed great potential in predicting HT in AIS patients. However, addressing methodological issues-such as inconsistent reference standards and limited external validation-is essential for the clinical implementation of these models.