Dual-energy CT angiography in detecting underlying causes of intracerebral hemorrhage: an observational cohort study Open Access 

Neuroradiology, published 10/25/24 | Estimated Read Time: 25 minutes

In a study of 206 ICH patients, dual-energy CT angiography (DECTA) demonstrated superior detection of underlying hemorrhage causes compared to conventional CTA, particularly for neoplasms (100% vs 57% detection), with fewer false positives (1.9% vs 2.9%) and false negatives (1.0% vs 5.8%). While the study's observational design and lack of standardized follow-up imaging limit definitive sensitivity calculations, the findings suggest DECTA could be a valuable first-line diagnostic tool for detecting hemorrhagic tumors and venous pathologies in acute settings.

Pulmonary Embolism Detection Without Intravenous Contrast Using Electron Density and Z-effective Maps from Dual-Energy CT

Radiology Advances, published 10/24/24 | Estimated Read Time: 15 minutes

Promising new technique using dual-energy CT's electron density and Z-effective maps to detect pulmonary embolism without contrast media, achieving 85% sensitivity and 86.92% specificity in a study of 150 patients. While the findings suggest potential applications for patients with contrast contraindications, the study's limitations include its single-center design, relatively small sample size with low PE prevalence (13.33%), and lack of correlation with D-dimer values. Key strength: excellent intra- and interobserver agreement (κ ≥0.9) suggests reliable reproducibility across different experience levels.

Decision Fusion Model for Predicting Microvascular Invasion in Hepatocellular Carcinoma Based on Multi-MR Habitat Imaging and Machine-Learning Classifiers Open Access 

Academic Radiology, published 10/28/24 | Estimated Read Time: 25 minutes

In a study of 300 hepatocellular carcinoma patients, researchers have developed a novel decision fusion model that achieved an AUC of 0.808 (95% CI: 0.714-0.902) in predicting microvascular invasion by combining clinical features, radiomics, and deep learning analysis of tumor habitat imaging within a 10 mm peritumoral region. While the model demonstrates promising predictive power through logistic regression classification (p > 0.050 on Hosmer-Lemeshow test indicating good calibration), and shows improved performance over single-modality approaches, the study's single-institution design (210 training/90 test split) and lack of external validation suggest the need for broader multi-center testing before clinical implementation.

Impact of Patient-reported Symptom Information on the Interpretation of MRI of the Lumbar Spine Open Access

Radiology, published 10/29/24 | Estimated Read Time: 20 minutes

Providing radiologists with patient-reported symptom information before lumbar spine MRI interpretation dramatically improves diagnostic accuracy. When armed with pre-MRI questionnaire results, radiologists achieved nearly perfect agreement with spine specialists (κ = 0.82-0.88) compared to only fair-to-moderate agreement without symptom information (κ = 0.26-0.51, p<.001), and reported significantly higher diagnostic confidence (80.4 vs 60.5 out of 100, p<.001). While the study was limited to fellowship-trained musculoskeletal radiologists at a single institution and lacked randomization, its findings suggest that implementing pre-MRI symptom questionnaires could substantially improve radiologists' ability to distinguish clinically relevant findings from incidental abnormalities in lumbar spine imaging.

Impact of non-contrast-enhanced imaging input sequences on the generation of virtual contrast-enhanced breast MRI scans using neural network Open Access

European Radiology, published 10/25/24 | Estimated Read Time: 15 minutes

Virtual contrast-enhanced (vCE) breast MRI demonstrates that combining T1-weighted imaging with multi-b-value diffusion-weighted imaging (DWI) including ultra-high b-values (b=1500 s/mm²) significantly improves the detection and visualization of breast lesions without requiring gadolinium contrast. In this single-center study of 1,064 breast MRI scans, the researchers found that while T2-weighted sequences improved overall image quality, they did not significantly enhance lesion detection capabilities. Key limitation: 6.5-8.6% of lesions showed minimal enhancement in vCE images even with optimal input combinations, and the study's single-center design requires validation across different MRI vendors and field strengths before clinical implementation.

Beyond MELD Score: Association of Machine Learning-derived CT Body Composition with 90-Day Mortality Post Transjugular Intrahepatic Portosystemic Shunt Placement Open Access 

Cardiovascular and Interventional Radiology, published 10/29/24 | Estimated Read Time: 25 minutes

A study of 122 cirrhotic patients demonstrates that machine learning-derived CT body composition metrics, when combined with MELD scores, significantly improved the prediction of 90-day mortality after TIPS placement (AUC 0.84 vs 0.76, p=0.02). While traditional BMI measurements showed no correlation with mortality, lower skeletal muscle area, subcutaneous fat, and visceral fat were independently associated with higher mortality.

Safety and Effectiveness of Drug-Eluting Embolic Bronchial Arterial Chemoembolization for Lung Cancer: A Systematic Review and Meta-analysis

Journal of Vascular and Interventional Radiology, published 10/29/24 | Estimated Read Time: 25 minutes

Meta-analysis of 18 studies (681 patients) reveals that drug-eluting embolic bronchial arterial chemoembolization (DEE-BACE) shows promise for lung cancer treatment, with significantly higher 6-month objective response rates (50.3%) and disease control rates (71.7%) compared to conventional BACE. Though the 12.23-month median survival and manageable safety profile are promising, the predominantly retrospective data highlights the urgent need for large-scale, prospective trials to confirm DEE-BACE's role in lung cancer care.

Analysis of the efficacy of Prostatic Artery Embolization in the treatment of Benign Prostatic Hyperplasia

Abdominal Radiology, published 10/26/24 | Estimated Read Time: 15 minutes

A recent study analyzing 107 patients demonstrates that Prostatic Artery Embolization (PAE) shows promising long-term efficacy for BPH-related lower urinary tract symptoms, with clinical success rates of 80.6% at 1 year and 79.3% at 2 years. The research identified key predictors of better outcomes: prostate volumes >65ml (OR=1.023, P=0.018), adenomatous-dominant BPH (OR=3.871, P=0.030), preoperative catheterization (OR=4.097, P=0.028), and bilateral embolization (OR=4.948, P=0.018). While the study shows encouraging results with minimal complications, limitations include its single-center retrospective design and potential learning curve effects over the 8-year study period.

Microwave versus Radiofrequency Ablation in Treating Predominantly Solid Benign Thyroid Nodules: A Randomized Controlled Trial

Radiology, published 10/22/24 | Estimated Read Time: 25 minutes

Multicenter RCT comparing microwave ablation (MWA) and radiofrequency ablation (RFA) for predominantly solid benign thyroid nodules. Researchers found MWA was noninferior to RFA in achieving volume reduction at 2 years (80% vs 83%, mean difference -2.4%, P<.001). While both techniques showed comparable efficacy and similar complication profiles, MWA procedures used less power (33.8W vs 48.5W), required less energy (18.2kJ vs 24.6kJ), and were more cost-effective ($1480 vs $1904) compared to RFA. 

Prediction of hepatocellular carcinoma response to radiation segmentectomy using an MRI-based machine learning approach

Abdominal Radiology, published 10/26/24 | Estimated Read Time: 25 minutes

Combining MRI radiomics features with clinical parameters can predict hepatocellular carcinoma response to Y90 radiation segmentectomy with fair accuracy (AUC 0.736), significantly outperforming traditional clinical parameters alone (AUC 0.531) and AFP (AUC 0.632). While promising for improving patient selection and treatment planning, this single-center retrospective study of 154 patients requires external validation before clinical implementation.

Comparing Commercial and Open-Source Large Language Models for Labeling Chest Radiograph Reports Open Access 

Radiology, published 10/29/24 | Estimated Read Time: 20 minutes

In a comparison of AI language models for radiology report analysis, researchers found that open-source models like Llama 2-70B achieved remarkably similar performance to GPT-4 in classifying chest X-ray findings, with micro F1 scores of 0.97 vs 0.98 (p>0.99) using zero-shot prompting on the ImaGenome dataset. While GPT-4 maintained slight superiority in some tasks, the study demonstrates that privacy-preserving, cost-effective open-source models could serve as viable alternatives for clinical report structuring, though limitations included potential prompt engineering bias favoring GPT-4 and significant class imbalances in the datasets. This research opens new possibilities for healthcare institutions seeking to automate report analysis while maintaining data privacy and controlling costs.

Large language models for structured reporting in radiology: past, present, and future Open Access 

European Radiology, published 10/23/24 | Estimated Read Time: 25 minutes

Large Language Models (LLMs) like GPT-4 can transform structured reporting in radiology, with studies showing up to 100% accuracy in template matching and report structuring. While current research is limited to just 10 studies primarily using GPT-3.5 and GPT-4, all reported promising results including successful multilingual applications. However, significant challenges remain - 27% of reports contained hallucinations in one study, 19% had misinterpretations, and there are major hurdles with regulatory compliance and the need for transparent algorithms and training data before clinical implementation.

Impact of deep Learning-enhanced contrast on diagnostic accuracy in stroke CT angiography Open Access 

European Journal of Radiology, published 10/28/24 | Estimated Read Time: 25 minutes

Single-center study of 102 stroke patients, deep learning-enhanced CT angiography (DLe-CTA) significantly improved detection of vessel occlusions in poorly contrasted scans, achieving 94% sensitivity compared to 82% with conventional CTA (p<0.001). While the technology shows particular promise for visualizing distal medium vessel occlusions, which are increasingly targeted for mechanical thrombectomy, the study's retrospective nature and single-center design warrant further validation across multiple institutions and scanner types.

WAW-TACE: A Hepatocellular Carcinoma Multiphase CT Dataset with Segmentations, Radiomics Features, and Clinical Data Open Access 

Radiology: Artificial Intelligence, published 10/23/24 | Estimated Read Time: 15 minutes

WAW-TACE is a comprehensive public dataset featuring 233 hepatocellular carcinoma patients (median age 66 years, 79.4% male) treated with TACE, including baseline multiphase CT scans, 377 hand-crafted tumor masks, and automated segmentations of 104 internal organs per patient. The dataset captures diverse disease presentations (63.9% single lesion, 36.1% multiple lesions) with robust outcome data (median overall survival 27.3 months, median progression-free survival 17.7 months), and includes 3,339 radiomic features per patient extracted using PyRadiomics. 

AI-derived Tumor Volume from Multiparametric MRI and Outcomes in Localized Prostate Cancer Open Access

Radiology, published 10/29/24 | Estimated Read Time: 25 minutes

Study of 732 prostate cancer patients demonstrates that AI-derived tumor volume measurements from multiparametric MRI can independently predict metastasis risk in both radiation therapy and radical prostatectomy patients (AHR 1.09, p=.001 and AHR 1.22, p=.001 respectively). The AI-based volume measurements outperformed traditional NCCN risk categories for predicting 7-year metastasis in radiation therapy patients (AUC 0.84 vs 0.74, p=.02), though the algorithm's lesion-level detection sensitivity ranged from 65-70% and required endorectal coil imaging, highlighting the need for further validation across different imaging protocols and institutions before clinical implementation.

Upcoming Conferences and Approaching Abstract Submission Deadlines

If you have any comments, questions, or feedback, please respond to this email!

Thank you,

Emily Barnard