Use Case Artificial Intelligence
The Eurostars project ImageREPORT is an innovative system is to improve radiology and the result of the combination of an automated medical image analysis and structured reporting. This project is founded upon the cooperation between Smart Reporting, Radboud UMC and Thirona.
Current practice for radiologists is to manually analyse medical images and write a text report, which is very time-consuming, and can lead to user-driven errors and inconsistent reports. Previous CAD software has not reached a sufficient accuracy for widespread implementation in clinical practice. ImageREPORT now significantly improves CAD accuracy through deep learning-based technology, while structured reporting improves communication and facilitates implementation in clinical practice. There is a growing awareness that structured reporting improves intra-hospital communication and confidence in clinical decisions by referring physicians. In addition, Computer-Aided Detection (CAD) methods, which automatically analyse and interpret large sets of medical images, increase detection accuracy and reduce interpretation time. However, previously developed CAD solutions show a high rate of false positive readings (up to 40%), severely hampering clinical usefulness. Recent developments in deep learning technology enabled development of improved CAD software, equalling experienced radiologists with respect to report quality. Radiology departments encounter an increasing volume of medical imaging data, particularly since novel high resolution imaging modalities, like helical CT scanners, result in a large number of images that must be manually inspected. Between 1999 and 2010, this led to a 10-fold increase in the number of images that need to be interpreted, with the number of imaging procedures increasing at a twofold rate in the same period.
The developed platform incorporates highly accurate CAD algorithms for automated analysis of chest CT-scans and chest X-rays (40% of the work load for the average radiology department) using deep convolutional neural networks, a type of deep learning technology. These results feed into a complementary reporting structure on the platform to ensure a seamless integration. In a nutshell, the project will result in an integrated radiology image analysis and reporting system that can be installed locally in hospital radiology departments.