An Integrated System For Radiology Reporting

Informatics has the potential to revolutionise healthcare practice, by allowing information on individual patients to be shared seamlessly and by facilitating the systematisation of medical knowledge. All of this assumes that the relevant information can be captured efficiently and represented unambiguously. In radiology, great advances have been made in image acquisition, but methods for capturing and representing the semantics of radiology reports (the main output of a radiology department) have received little attention. Standard practice is for the radiologist to dictate a textual report as s/he assesses a patient’s images. The images are usually presented on film - even when the original acquisition was digital - and the recorded report is subsequently typed up for inclusion in the patient’s notes. Thus the information is captured via an impoverished interface and recorded in a form where the semantic content is inaccessible. This means that the huge investment made by the NHS in imaging equipment and imaging investigations is used inefficiently in the management of individual patients, and is virtually unusable for data-mining and knowledge systemisation.

The reason for this state of affairs is that the problem of capturing and representing the information a radiologist needs to convey is very difficult. Existing attempts to provide electronic reporting (including products offered by major manufacturers) have been singularly unsuccessful - none provides a truly integrated solution to reporting, and they all severely reduce the radiologist’s (short-term) productivity. A practical system needs to allow routine tasks to be performed very rapidly, whilst providing sophisticated facilities for dealing with unusual cases. If run-of-the-mill work cannot be completed at least as quickly as is possible with existing non-informatics solutions, systems are not used and the potential for informatics to add value is never realised.

We believe the solution is to develop more intelligent systems that understand and anticipate what the user is trying to do, reducing the need for interaction. The key is to integrate all aspects of reporting in a single system, with a high-bandwidth user-interface and direct links to other health informatics systems, particularly the electronic patient record. If the system is to ‘understand’ the user, and allow natural interaction, this requires the integration of several technologies: speech input, automated image analysis, medical terminology analysis, and knowledge representation. This is an ambitious undertaking, and we do not suggest that it can be completed in the project we propose. There are, however, important research issues that can only be addressed by starting to tackle the whole problem. We are perhaps unique in the UK in having all the relevant expertise. The proposed project involves a collaboration between the University Division of Imaging Science and Biomedical Engineering, and the Department of Computer Science, bringing together expertise in radiological practice, medical image analysis, process modelling, description logics, and medical ontologies. We plan to develop and assess an initial framework for an integrated reporting system, solving some of the research problems that we can already see, and defining a longer-term research agenda by working together. The specific objectives of the project are to:

model the radiology reporting process;

develop a formal representation of radiology reporting terminology;

extend existing methods of model-based image interpretation to be independent of imaging modality;

develop a prototype system integrating speech recognition, medical terminology, and automated image analysis technologies;

evaluate the prototype system.