|ECR 2018 / C-2810|
|Multimedia structured reporting employing natural language processing to improve the efficiency of data entry|
The reporting system is deployed in clinical trials at an academic medical center. The NLP engine labels image findings with a high-degree of accuracy, but when it fails to find an exact match, the system presents the radiologist with a list of potential matches from which the radiologist can perform final editing.
Initial labeling failures were attributed to incompleteness of the ontology which were corrected during development. For example, the phrase "left gastric lymph node" yielded "stomach" and "lymph node," but after the complete phrase of "left gastric lymph node" was incorporated into the metadata, the system delivered the desired result.
Another ontology-related issue has been the need to incorporate adjectives and synonyms into the ontology. For example, the term "gastric" is an adjective of "stomach," and the NLP engine and ontology had to account for both relating to the central concept of "stomach" (SNOMED CT code 69695003).
Some concepts have multiple synonyms, and those also had to be taken into account when constructing the metadata. For example, "normal" can be inferred by several spoken terms and phrases including "unremarkable," "no significant abnormality," and "negative."
Refinement of the NLP algorithms and ontology are ongoing aims. To date the reporting system's metadata that is referenced to SNOMED CT terminology comprises over 14,400 radiological observations and diagnoses.
A benefit of creating structured reports using standardized language is that the database of image findings can be mined for healthcare outcomes (figure 9). In this example, the database can be queried to ask what percentage of patients with a certain disease treated with a specific regimen have responded to therapy, and the system can produce the answer using the linked structured data elements.
An online demonstration of the system can be found at:
Thematically related posters
ECR 2018 / C-2352
An electronic system for submitting and reviewing resident schedule requests increases resident satisfaction and residency program administrative efficiency
ECR 2018 / C-2287
Building a Better Data Lake: Enterprise Data Discovery within a Hospital PACS Environment
ECR 2018 / C-2510
Automating the Calculation of Clinical Outcomes with Multimedia Structured Reporting to Satisfy Pay-For-Performance Requirements