Using ontologies and data analysis techniques, researchers have helped clean up RadLex, the RSNA’s pilot radiological lexicon project, according to two presentations at the recent RSNA meeting.
Using ontologies and data analysis techniques, researchers have helped clean up RadLex, the RSNA's pilot radiological lexicon project, according to two presentations at the recent RSNA meeting.
Maintaining and curating RadLex is difficult, said Dr. Daniel L. Rubin, a clinical assistant professor at Stanford University Medical Center. RadLex resides in a flat file format that is friendly to computers but unfriendly to people.
Rubin outlined how ontologies, which describe concepts and relationships in a way that is comprehensible to humans and usable by computers, can help identify omissions and redundancies in RadLex.
Using a program developed at Stanford called Protégé, Rubin and colleagues mapped RadLex terms and attributes to ontology classes and relationships. This created a graphical, user-friendly way to browse through the terms.
The ontology model allowed researchers to determine that 23 of the 1326 RadLex concepts are synonyms. They also found 17 duplicate terms, with 15 of them in separate ontological subtrees. Because the terms resided in entirely different subtrees, they were probably not redundant, according to Rubin. The other two cases were in the same subtrees and probably did represent redundant terms.
Dr. Dirk Marwede at the University Hospital in Leipzig, Germany, presented the results of his analysis comparing radiological terms found in thoracic CT reports with terms used in RadLex's draft lexicon for thoracic radiology.
Marwede and colleagues randomly selected 250 thoracic CT reports from a database. They then extracted and compared the appropriate terms from the reports with terms found in RadLex. They found 512 relevant terms, 68% of which were found in the RadLex term categories Findings and Anatomic Location. Marwede reported that the visual features and anatomic location findings were the most frequently used RadLex categories.
The study found that 164 terms did not appear in RadLex. While most terms used in thoracic CT reports are already contained in RadLex, relationships defined in the lexicon need to be more specific, and indexing report content needs well-defined encoding rules, Marwede said.
Because the development of RadLex is an ongoing process, this type of research and refinement of the lexicon is exactly what RadLex needs, said Dr. Donald P. Harrington, chair of radiology at Stony Brook University School of Medicine.
The complete RadLex lexicon should be available by the 2006 RSNA meeting.
MRI-Based AI Radiomics Model Offers 'Robust' Prediction of Perineural Invasion in Prostate Cancer
July 26th 2024A model that combines MRI-based deep learning radiomics and clinical factors demonstrated an 84.8 percent ROC AUC and a 92.6 percent precision-recall AUC for predicting perineural invasion in prostate cancer cases.
Breast MRI Study Examines Common Factors with False Negatives and False Positives
July 24th 2024The absence of ipsilateral breast hypervascularity is three times more likely to be associated with false-negative findings on breast MRI and non-mass enhancement lesions have a 4.5-fold likelihood of being linked to false-positive results, according to new research.
Can Polyenergetic Reconstruction Help Resolve Streak Artifacts in Photon Counting CT?
July 22nd 2024New research looking at photon-counting computed tomography (PCCT) demonstrated significantly reduced variation and tracheal air density attenuation with polyenergetic reconstruction in contrast to monoenergetic reconstruction on chest CT.