Here we summarise a recent review article, published in Neuroimaging Clinics, that explored the additive value of artificial intelligence (AI) for process and workflow improvement in radiology.
Natural Language Processing
A lot of attention has been paid to the use of AI for image analysis. However, natural language processing (NLP), a form of AI, is also important for developing applications and tools that enhance radiology practice. Experts can utilise NLP to process unstructured data and make it interpretable to computers and algorithms. Like other machine learning algorithms, NLP classifiers rely on the quality of the dataset – a key feature being its size.
AI in the workflow
There are many important applications for AI in the workflow. These include image analysis for diagnostic and interpretation tasks, workflow optimisation and support for quality and safety and operational efficiency. In radiology, advances in machine learning (ML) approaches, in particular, have garnered a lot of interest. There is great potential in the application of ML for image analysis diagnostic decision support tasks and precision medicine. Nevertheless, the potential applications of AI in radiology go beyond just image analysis and can enhance all levels of the radiology workflow and practice. Here, we summarise the various applications of AI in the radiology workflow:
- Pre-image acquisition: AI could optimise the process of radiology scan ordering and reduce ordering of unnecessary scans. It could also automate retrieval of clinical data from the EHR, improving the quality and efficiency of clinical decision support software in the future.
- During image acquisition: AI has the potential to improve the radiology scan acquisition process. This includes as assistive technology for improving radiology technologists and enabling increased scan quality and acquisition efficiency. For example, AI has the potential to assist in patient positioning during CT scans but also to optimise CT dose.
- Post-image acquisition: AI could help orchestrate a more efficient and optimal workflow and image routing. It can enhance existing informatic solutions, image management and analytics. For example, a widely used application of AI is reading list prioritisation i.e. flagging critical findings. AI can also play a role in improving quality assurance, enhanced patient safety and also educational purposes.
- Other applications: AI can be used for image-based prediction of treatment response or prognosis. This is largely a focus in oncology. Other nondiagnostic tasks include improving examination scheduling and using predictive analytics to ensure optimal manpower needs.
While the focus of AI applications has largely been on diagnostic support and precision medicine, the other important applications of AI should not be forgotten. This includes applications related to process improvement and operational efficiency. These applications have the potential to have a significant impact on healthcare delivery and likely easier adoption. ML-based algorithms are not currently well integrated into clinical ecosystems. The authors noted that a robust informatics infrastructure and seamless data integration are essential to fully exploit these applications. If done correctly, these applications could not only improve healthcare system efficiency and sustainability, but also improve the quality of patient care.
Image credit: By freepik – www.freepik.com