Using artificial intelligence to help diagnose disease

Areas where radiological imaging is frequently needed may benefit from using artificial intelligence (AI) to help with diagnosis, suggest findings published by Bristol BRC researchers in BMJ Open. Future research should focus on how the technology could be used by healthcare professionals in clinical settings.

Healthcare professionals use digital technologies on a day-to-day basis throughout the health and care system. Researchers and healthcare professionals could use the information gathered with the help of these technologies to train AI models in pattern recognition. They could then use these trained AI models as diagnostic tools in, amongst others, radiological imaging.

Radiology or radiological imaging is a medical discipline that uses images of the human body to diagnose diseases and guide how they are treated. For example, it can be used by surgeons to find out whether a patient’s cancer has returned or whether a tumour is malignant or benign.

The research team reviewed 15 studies to explore and appraise how AI models had been used to identify abnormalities on images of the abdomen and pelvis. Most of the included studies used AI models to diagnose advanced or recurrent cancer. Four studies focussed on classifying whether an abnormality was benign or malignant and only four studies compared the AI model with human performance.

Researchers found that AI models were used in very diverse ways across different surgical specialities. They found that the accuracy of AI models asked to diagnose advanced or recurrent cancers could support healthcare professionals while they make decisions about potential diagnoses and treatments.

However, the study team also identified some issues associated with how information about AI use was reported in a surgical setting. This included a lack of data on how the models were trained and the reporting of studies was unstandardised, which in turn creates problems for studies assessing the accuracy of AI use in healthcare.

Dr George Fowler, lead author, said:

“AI could have great potential in clinical applications. AI could support clinicians in diagnostic imaging to make care more efficacious and effective.

“Examples include the interpretation of medical images for diagnostic, prognostic, surveillance and management decisions, which otherwise rely on a limited number of interpreters and human resources.”

Dr Rhiannon Macefield, a member of the research team, said:

“This technology could potentially reduce the clinical workload. However, our review found many of the AI studies were retrospective and/or proof-of-concept. While this may be appropriate for early phase surgical research, future efforts should evaluate the role of AI in a clinical setting. In particular targeting areas where radiological expertise is in high demand or to support diagnoses where the images are difficult to interpret.”