Background: Australian cancer registries collect a range of data from hospitals and pathology laboratories but information on staging and recurrence remains an acknowledged void in population-based data for cancer control. Â Diagnostic imaging reports provide a rich source of information pertaining to stage and recurrence. Natural Language Processing (NLP) has potential to capture these data from imaging report electronic repositories.
Aim: To develop NLP algorithms for cancer registries to capture and extract information on stage and recurrence in real time from diagnostic imaging report repositories..
Methods: 1. Develop and trial NLP algorithms to capture stage and recurrence from imaging reports. 2. Compare NLP data capture with that of clinical coders. 3. Incorporate NLP derived information with existing registry data to evaluate the capacity of the product to deliver staging and recurrence at the population level.
Results: Our NLP trials have achieved 100% sensitivity and 97% specificity for identifying reports relevant to cancer. Stage at diagnosis was successfully extracted for 97% of lung cancer cases in sample data set .
Discussion: Use ofNLP technology by registries to interrogate diagnostic imaging databases can significantly contribute to bridging the void in population-based data on cancer stage and recurrence with minimal burden on health services and in a timely fashion.