CT Perfusion imaging as prognostic factor for outcome of lacunar stroke Open Access 

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

New research suggests global Cerebral Blood Flow (gCBF) measured by CT Perfusion could be a valuable prognostic tool for lacunar stroke outcomes. In a study of 162 patients, those with gCBF below 40 ml/100mg/min showed significantly higher rates of early neurological deterioration (37.8% vs 3.1%) within 48 hours. 

The Ovarian-Adnexal Reporting and Data System (O-RADS) US Score Effect on Surgical Resection Rate Open Access 

Radiology, published 10/8/24 | Estimated Read Time: 15 minutes

Nearly half of surgically removed ovarian lesions could have been managed conservatively using the Ovarian-Adnexal Imaging Reporting and Data System (O-RADS) US version 2022 risk score. The research found that 42% of excised lesions met O-RADS US 2 criteria, with a malignancy rate of less than 1% in this category. The study demonstrated high sensitivity (94%) and negative predictive value (98%) for predicting malignancy using O-RADS US 4 as the threshold.

Radiomics Biomarkers to Predict Checkpoint Inhibitor Pneumonitis in Non-small Cell Lung Cancer

Academic Radiology, published 10/11/24 | Estimated Read Time: 20 minutes

A radiomics-clinical model has been developed to predict checkpoint inhibitor pneumonitis (CIP) in non-small cell lung cancer patients before immunotherapy. By combining CT-based radiomics features with clinical risk factors, the model achieved impressive AUCs of 0.935, 0.905, and 0.923 in training, validation, and external validation cohorts, respectively. While the study's relatively small sample size is a limitation, this innovative approach could significantly improve treatment selection and patient outcomes. 

Mri-only breast cancers are less aggressive than cancers Identifiable on conventional imaging 

European Journal of Radiology, published 10/11/24 | Estimated Read Time: 20 minutes

New study suggests that breast cancers detected only on MRI are less aggressive than those visible on conventional imaging. The research found MRI-only lesions were significantly smaller, had lower Ki-67 levels, and were more likely to be pure DCIS (30.4% vs 5.2%) compared to lesions detected by second-look ultrasound. Additionally, MRI-only cases had a higher rate of negative axillary lymph nodes (87.2% vs 68.3%). These findings may support follow-up rather than immediate biopsy for some MRI-only lesions, but the retrospective design and potential influence of reader experience on ultrasound detection rates are limitations of the study. 

MRI-based Zero Echo Time and Black Bone Pseudo-CT Compared with Whole-Body CT to Detect Osteolytic Lesions in Multiple Myeloma Open Access 

Radiology, published 10/8/24 | Estimated Read Time: 25 minutes

Possible radiation-free alternative to CT for detecting osteolytic lesions in multiple myeloma by using MRI-based zero echo time (ZTE) and black bone (BB) pseudo-CT sequences. Both sequences demonstrated high diagnostic accuracy (98% for experienced readers), with BB outperforming ZTE in the spine. The study was limited to the lumbar spine, pelvis, and proximal femurs, warranting further research for whole-body application. 

Streamlining Radioembolization without Lung Shunt Estimation versus Regular Radioembolization in Patients with Hepatocellular Carcinoma within Milan Criteria 

Journal of Vascular and Interventional Radiology, published 10/12/24 | Estimated Read Time: 20 minutes

Retrospective study suggests that streamlined radioembolization (S-TARE) without lung shunt estimation is as safe and effective as regular radioembolization (R-TARE) for hepatocellular carcinoma within Milan criteria. The research found no significant differences in complete response rates (S-TARE: 90.3%, R-TARE: 92.1%, p=1.000) or time to progression, with comparable safety profiles and no reported radiation pneumonitis. While promising, the study's retrospective nature and limited sample size, especially for resin microspheres, warrant further investigation.

Developing a Clinical Service in Interventional Radiology: Results from the 2024 CIRSE Clinical Practice Survey Open Access 

Cardiovascular and Interventional Radiology, published 10/9/24 | Estimated Read Time: 15 minutes

A CIRSE survey with 520 respondents reveals encouraging progress in clinical practice development for Interventional Radiology, with 62.1% of respondents accepting patients as the primary consultant and 49.6% operating IR outpatient clinics. The study advocates for a step-wise approach to developing clinical services, starting with ward rounds (performed by 45% of respondents) and progressively increasing responsibility. While 75% of IRs feel confident in their clinical performance, only 46% received dedicated clinical practice education during training, highlighting a need for more focused education. However, the bias towards large academic centers limits the generalizability of these findings to all IR practices. 

Efficacy of radiotherapy combined with hepatic arterial infusion chemotherapy, TKI and ICI for hepatocellular carcinoma with portal vein tumor thrombus: a retrospective cohort study

Abdominal Radiology, published 10/11/24 | Estimated Read Time: 20 minutes

Combination therapy of radiotherapy, hepatic arterial infusion chemotherapy, TKI, and ICI shows promising results for hepatocellular carcinoma patients with portal vein tumor thrombus. This retrospective study found significantly improved progression-free survival (HR 0.459, P=0.008), overall survival (HR 0.420, P=0.024), and tumor response rates compared to HAIC+TKI+ICI alone, without increased adverse events. However, the small sample size and lack of randomization warrant further investigation.

A Comparison of Post-Procedural Hemoglobin in Catheter-Directed Thrombolysis vs. Large-Bore Aspiration Thrombectomy for Acute Pulmonary Embolism

Journal of Vascular and Interventional Radiology, published 10/12/24 | Estimated Read Time: 15 minutes

No significant difference was found for immediate post-procedural hemoglobin changes(p=0.098) between catheter-directed thrombolysis (CDL, n=166) and large-bore aspiration thrombectomy (LBAT, n=105) for acute pulmonary embolism. However, LBAT without autotransfusion showed a greater hemoglobin decrease at the 7-day nadir (p=0.008). While LBAT groups had more minor hemorrhagic events (p=0.010), there was no significant difference in moderate/severe events or transfusion rates, suggesting treatment choice shouldn't solely be based on expected blood loss.

The Influence of Pre-ablation Embolization Particle Size on the Size of the Microwave Ablation Zone in a Porcine Orthotopic Renal Tumor Model

Journal of Vascular and Interventional Radiology, published 10/10/24 | Estimated Read Time: 20 minutes

Researchers explore combining arterial embolization with microwave ablation to treat renal tumors in a porcine model. Pre-ablation embolization using smaller particles (40-100 μm) significantly increased ablation zone volume by up to 66% compared to ablation alone. While promising for expanding treatment options for larger renal tumors, the study was limited by small sample size and use of a novel animal model that may not fully replicate human renal cell carcinoma vascularity.

Automatic structuring of radiology reports with on-premise open-source large language models Open Access 

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

Researchers have developed a method using an open-source large language model (Llama-2-70B) to automatically convert free-text radiology reports into structured formats with human-level accuracy. The study, conducted on chest radiograph reports in English and German, achieved Matthews correlation coefficients of 0.75 and 0.66 respectively, demonstrating the potential to enhance report processing efficiency while preserving patient privacy. However, performance varied across languages and imaging findings, highlighting the need for further evaluation in larger clinical cohorts.

Enhancing Radiologists’ Performance in Detecting Cerebral Aneurysms Using a Deep Learning Model: A Multicenter Study Open Access 

Academic Radiology, published 10/14/24 | Estimated Read Time: 15 minutes

Multicenter study shows that a deep learning model significantly improves radiologists' performance in detecting cerebral aneurysms on CT angiography, especially for small aneurysms (< 3mm) and less experienced readers. The AI assistance increased sensitivity from 68.9% to 81.6% for junior radiologists and from 72.4% to 83.5% for senior radiologists, while also reducing image interpretation time by 37.2% and post-processing time by 90.8%. However, the study lacked DSA validation and excluded patients with prior vascular interventions.

Assessment of a fully-automated diagnostic AI software in prostate MRI: Clinical evaluation and histopathological correlation

European Journal of Radiology, published 10/14/24 | Estimated Read Time: 20 minutes

A new study evaluates the performance of mdprostate, an AI-driven software tool, in detecting and grading prostate lesions on MRI. The AI achieved 100% sensitivity for ruling out prostate cancer at PI-RADS ≥2 and demonstrated comparable performance to expert radiologists using PI-RADS v2.1 for clinically significant cancer detection (85.5% sensitivity, 63.2% specificity at PI-RADS ≥4). While promising, the study's retrospective design and single-center cohort warrant further validation in larger, prospective trials.

Retrieval-Augmented Generation for Large Language Models in Radiology: Another Leap Forward in Board Examination Performance Open Access 

Radiology, published 10/8/24 | Estimated Read Time: 15 minutes

Retrieval-augmented generation (RAG) enhances large language model performance in radiology. Using Perplexity Pro with GPT-4 Turbo, researchers achieved a remarkable 90% accuracy on radiology board-style examination questions, significantly outperforming ChatGPT-4's 79% (p=0.001). While this leap forward demonstrates RAG's potential to reduce hallucinations and improve domain-specific knowledge, the study's limitations include the absence of image-based questions and relatively small subgroup sizes for analysis.

Boosting Deep Learning for Interpretable Brain MRI Lesion Detection through the Integration of Radiology Report Information Open Access 

Radiology: Artificial Intelligence, published 10/9/24 | Estimated Read Time: 15 minutes

A deep learning model incorporating radiology report-derived textual features outperformed a basic model in diagnosing 14 brain diseases and normal brains on MRI, achieving higher AUCs (0.93 vs 0.85) and producing more interpretable attention maps. The report-guided model showed enhanced generalizability across different medical centers and improved radiologists' diagnostic performance. However, the study was limited by its retrospective nature and focus on specific brain pathologies, warranting further research on a broader spectrum of conditions.

Upcoming Conferences and Approaching Abstract Submission Deadlines

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

Thank you,

Emily Barnard