Researchers from the faculties of Engineering and IT at Monash University have come up with an AI algorithm that can judge another AI algorithm’s annotation or label in a medical scan, mimicking the process of seeking a second opinion.
They created a dual-view AI system where one part labels medical images while the other judges the quality of the AI-generated labelled scans by benchmarking them against radiologist-provided labelled scans. Researchers used 10% labelled data from three publicly accessible medical datasets.
Based on findings published in the journal Nature Machine Intelligence, the AI system achieved a 3% improvement “compared to most recent state-of-the-art approach under identical conditions.”
“It demonstrates remarkable performance even with limited annotations, unlike algorithms that rely on large volumes of annotated data,” said principal researcher Himashi Peiris, a PhD candidate from the Faculty of Engineering.
WHY IT MATTERS
The main goal of the research was to address the limited availability of human-annotated or labelled medical images by using a competitive learning approach against unlabelled data.
A traditional method of labelling medical scans by hand can be time-consuming, prone to errors, and relies on an individual’s subjective interpretation. It can also extend waiting periods for patients seeking treatments.
Meanwhile, large-scale annotated medical image datasets are often limited as manual annotation requires significant time, effort, and expertise.
The algorithm in the Monash research allows multiple AI models to “leverage advantages from labelled and unlabelled data, and learn from each other’s predictions to help improve overall accuracy.” It also allows them to “make more informed decisions, validate their initial assessments, and uncover more accurate diagnoses and treatment decisions.”
The researchers are now working to expand their AI system to work with different types of medical images and develop a dedicated end-to-end product for practices.
THE LARGER TREND
One of the terrific use cases of AI in healthcare is supporting clinician decisions and supplementing clinical diagnoses. A popular example is IBM’s Watson which uses various AIs to sort through information and provide clinical insights and recommendations for personalised treatments. The Watson system has commercialised applications for genomics, drug discovery, health care management, and oncology.