The objective of this project is to build a commercially-viable automated system to identify records required for population-based cancer surveillance and extract from them specific data elements and their values. The system is to express the elements and values in standard nomenclatures and transmit them to defined destinations. A prototype system using artificial intelligence techniques has been successfully developed to extract reports of CNS neoplasms from CT and MRI reports. During this project that prototype wi1J be extended to receive reports for a second cancer site and to incorporate MRI-F and PET reports. The revised prototype will be implemented in four production environments, transmitting to three central registries. The reports wilt be processed by the software and also manually reviewed by a team of expert CTR's. Results will be compared and the software will be improved to achieve sensitivity and specificity goals in the 98-99% range. The benefits to central cancer registries will be measured in terms of case ascertainment, level of effort and timeliness. A marketing effort will be launched to determine the commercial viability and identify customers for the software. Post Phase II, other lexicons will be added to process other cancer sites found through MRI/CT/PET/F-MRI techniques.
