A review of our project with the Victorian and NSW Cancer Registries has just been provided to us by Cancer Australia. Here are some of the assessments:
The final version of the case ascertainment classifier successfully achieved the revised target for sensitivity of 99%, and the initial target for specificity of 96%, to correctly identify reports for cancer from one diagnostic imaging service (Pilot Site 1). The development of this classifier demonstrates the potential for natural language processing technology to automate the selection of diagnostic imaging reports for cancer.
On the extraction engine, Cancer Australia say:
The results demonstrate the potential of natural language processing technology to automate the extraction of staging and recurrence information from diagnostic imaging reports.
On the processing pipeline that we built, Cancer Australia say:
The pipeline was found to function perfectly. All reports were complete and complied for the minimum dataset format.
The development of the TumourTExtract pipeline has demonstrated its capacity to automate the ascertainment of diagnostic imaging reports, secure transfer of reports for cancer to the registry, to extract a minimum dataset and to output the resulting information in a standard format (XML batch file). This technology has been successfully implemented with minimal burden on diagnostic imaging services’ day-to-day operations.
Further, at the conclusion of the project, the TumourTExtract pipeline was delivering reports from the diagnostic imaging services to the Victorian Cancer Registry in near ‘real-time’.