Summary:
UCLA researchers have developed innovative software that streamlines radiological data analysis to enhance patient outcomes, evaluate radiologist performance, and assess AI algorithms in a way that mirrors radiologist decision-making.
Background:
Radiological imaging is critical to disease diagnosis and treatment planning for a wide array of disease states. Traditional methods of determining diagnostic accuracy and radiologist performance include peer review systems, which rely on retrospective analysis of diagnostic images by other radiologists using simple scoring systems. Accordingly, these methods do not capture the significance of errors and do not provide metrics to assess diagnostic performance and accuracy across varying subspecialities. Other structured feedback systems are limited to specific diseases and diagnostic contexts and cannot be used across radiology departments. These approaches are also limited to small datasets and cannot be used to map radiology findings and outcomes from large hospitals and departments. Novel AI algorithms have been widely developed across imaging modalities, enhancing diagnostic capabilities and streamlining clinician workflow. However, trust in AI software remains a challenge due to concerns about training data requirements and accuracy. There is still an unmet need for a scalable, broadly applicable method to analyze radiological findings and support AI algorithm development.
Innovation:
UCLA researchers from the Department of Radiological Sciences have developed a data-driven software application to assess and improve diagnostic accuracy of radiological findings and improve AI algorithm development. This software integrates electronic medical records with radiology information to automatically extract diagnostic statements from reports using natural language processing. This information is then classified into general and disease specific categories, which can be used to assess the precision of diagnosis data from various sources, such as pathology, surgery, or laboratory tests. The reported technology can also be used to identify and categorize radiological findings from large diagnostic datasets, enabling both real-time and retrospective assessment of radiologists’ diagnostic performance. Additionally, the system features a scoring algorithm that quantifies the contribution of radiologic interpretation to patient care, emphasizing diagnostic specificity and actionable insights to support clinical decision-making. This innovation is adaptable to various diagnostic imaging settings, such as hospitals and clinics, while also driving the development of next-generation AI algorithms for medical image interpretation. By reducing diagnostic variability and error rate, this innovation has the potential to revolutionize radiological imaging by driving continuous improvements in diagnostic practices and results, and manage these same parameters in radiologic AI.
Potential Applications:
• Radiologist performance evaluation
• Quality assurance
• Integration with radiology software
• Medical education and training
• Clinical research
• Decision support for multidisciplinary care teams
Advantages:
• Diagnostic accuracy assessment
• Real-time and retrospective performance evaluation
• Data-driven scoring
• Improved concordance between imaging and precision diagnoses
• Reduction in diagnostic errors and variability
Development-To-Date:
First successful demonstration of the invention
Reference:
UCLA Case No. 2014-501
Patent Application:
Automated quality control of diagnostic radiology