Room: Exhibit Hall | Forum 7
Purpose: Structure naming conventions introduced in TG-263 are needed for data integrity of outcome data research but can be difficult to implement. We present an open source library for evaluating and aiding clinics for easier naming compliance
Methods: Using the Eclipse Scripting API, we developed a framework which evaluated structure labels and proposed the correct structure label if non-compliance was detected. The evaluator consisted of a series of regular expressions which could identify string patterns which did not obey guidelines. The recommendation engine created variations of each word, calculated the Levenshtein distance between a TG-263 dictionary and the violating word combinations, sorted the results by distance and proposed 5 suggested compliant labels. For testing, 252 structure labels were randomly selected from the database and these were input into the software. For each label, a physicist manually checked the compliance and reported if the computed suggestion was correct. Because the software recommended labels in order, the position of the recommendation was recorded from 1 to 5.
Results: 245 cases (97.2%) were correctly marked as compliant or non-compliant and 7 (2.7%) were incorrectly marked. 217 labels of the sample were non-compliant. When 5 suggested labels were proposed for the 217 non-compliant labels, 61 cases were found to not predict the correct label and 44 could not be determined by the physicist. For the remaining 112 cases, the correct label was recommended. The mean position of the correct recommendation was 1.49Â±0.75 out of the 5 labels suggested.
Conclusion: We have demonstrated an accurate system for detecting errors and recommending structure labels for TG-263 compliance. This open source resource has been published to the Eclipse Scripting API X (ESAPIX) library on github.com and is available for use.
Not Applicable / None Entered.