Autonomous large-scale findings analysis for effective diagnostic imaging

Purpose and goal

The goal of the project was to create an operation management tool geared towards imaging diagnostics. The tool should, with the aid of AI, analyze requests and reports in order to answer the question of whether we are doing the right things. Within the project we developed a tool that automatically can find the number of positive findings in a collection of reports. This statistic can be tracked over time and be used by operations management to monitor changes in reasons for performing an examination. Another use is monitoring that equal treatment is given to different patient populations.

Expected results and effects

The main result in the project is a system for finding frequency analysis aimed at being used as support in discussions regarding reasons for performing an examination. The project has also produced scientific publications. Methods developed for automatic analysis of reports was presented at the European Congress of Radiology 2017. The work has also been the basis for two scientific reports performed during doctors’ specialty training. Methods using transfer learning with language models, developed in the project, could serve as a baseline for future work within text analysis.

Planned approach and implementation

The project was a collaboration between Sectra, Uppsala Universitet and Röntgen Region Östergötland. In the early phases of the project Region Östergötland was the driving force, as we initially needed to define the use cases and annotate data. Development of algorithms and prototypes was mainly performed at Sectra. Sectra’s expertise and software development platform helped in making this part a success. There have been no major hurdles in the project, with is unusual. This might in part explain why the project ends a bit under budget.