Radiology in public focus

Fine-tuned LLMs Boost Error Detection in Radiology Reports  

Fine-tuned large language models (LLMs) greatly enhance error detection in radiology reports, according to a study published in Radiology.  

Senior study author Yifan Peng, PhD, from the Department of Population Health Sciences at Weill Cornell Medicine in New York, and colleagues evaluated fine-tuned LLMs for detecting errors in radiology reports during medical proofreading. A fine-tuned LLM is a pre-trained language model with additional training on smaller, targeted datasets. 

The model was tested on a dataset composed of two parts. The first consisted of 1,656 synthetic reports, including 828 error-free reports and 828 reports with errors. The second comprised 614 reports, including 307 error-free reports from MIMIC-CXR, a large, publicly available database of chest X-rays, and 307 synthetic reports with errors. 

The researchers found that the model outperformed both GPT-4 and BiomedBERT, a natural language processing tool for biomedical research, providing evidence that LLMS can assist in various types of error detection.  

“We are excited to keep exploring innovative strategies to enhance the reasoning capabilities of fine-tuned LLMs in medical proofreading tasks,” Dr. Peng said. “Our goal is to develop transparent and understandable models that radiologists can confidently trust and fully embrace.” 

Read the related RSNA News story.

Shorter MRI Exam Effectively Detects Cancer in Dense Breasts

Abbreviated breast MRI shortens exam time while retaining a high level of diagnostic accuracy of breast cancer in women with extremely dense breasts, according to an article published in Radiology. 

While breast MRI is recommended as an additional cancer screening method in women with extremely dense breasts, the added time and cost of the procedure has impacted its widespread application. 

To determine the minimal protocol necessary to detect breast cancer, study author Wouter B. Veldhuis, MD, PhD, an associate professor in the Department of Radiology at the University Medical Center at Utrecht University in the Netherlands and colleagues conducted a secondary analysis on a subset of data from Netherlands-based Dense Tissue and Early Breast Neoplasm Screening (DENSE) Trial. 

Seven radiologists, each with a minimum of 16 years of experience, read the same set of screening breast MRI exams with four different sequences added incrementally, resulting in a total of 2,072 reads.

The researchers found that the abbreviated MRI protocol had comparable sensitivity and specificity to the full protocol MRI, while also requiring an 80% shorten scan time and 50% shorter pooled reading times. The shorter procedure time has the added benefit of requiring less time lying in an MRI machine, decreasing patient discomfort.  

“Reducing the examination time and noise levels potentially improves patient experience,” Dr. Veldhuis said. “Moreover, shorter reading and scan times may allow implementation of MRI in national screening programs making it available for all women with extremely dense breasts.”

 Read the related RSNA News story. 

Special Report Highlights LLM Cybersecurity Threats in Radiology-review  

Researchers address the cybersecurity challenges of large language models (LLMs) and the practical implementation of security measures to prevent LLMs from being used maliciously in the health care system, according to a special report published in Radiology: Artificial Intelligence. 

An increasing number of health care providers are exploring ways to integrate LLMS into their daily workflows, including clinical decision support, patient data analysis and enhancing communications between physicians and patients. However, they can be vulnerable to security threats and malicious interference.  

Lead author Tugba Akinci D’Antonoli, MD, neuroradiology fellow in the Department of Diagnostic and Interventional Neuroradiology, University Hospital Basel, Switzerland, discussed the ways that LLMS are vulnerable to attacks and methods to safely integrate LLMS into health care.  

Implementing robust security measures and adherence to best practices from development to deployment are just some of the methods to minimize risks while protecting patient privacy.  

“The landscape is changing, and the potential for vulnerability might grow when LLMs are integrated into hospital systems,” Dr. D’Antonoli said. “There is increasing awareness, stronger regulations and active investment in cybersecurity infrastructure. So, while patients should stay informed, they can also be reassured that these risks are being taken seriously, and steps are being taken to protect their data. 

Read the related RSNA News story.