Monash University researchers have developed an innovative AI algorithm for medical imaging that mimics the process of seeking a second opinion. The algorithm utilizes both labeled and unlabeled data, achieving a 3% improvement compared to the latest state-of-the-art approaches. It significantly enhances medical image analysis, providing more accurate diagnoses and treatment decisions. The next step involves expanding its application to various medical image types and creating a practical tool for radiologists’ use.
Recent research from Monash University showcases a groundbreaking AI algorithm for medical imaging, emulating the process of seeking a second opinion. Published in Nature Machine Intelligence, this study addresses the scarcity of human-annotated medical images by employing an adversarial learning approach with unlabeled data.
The innovative co-training AI algorithm, developed by researchers from Monash University’s Faculties of Engineering and IT, represents a significant advancement in medical image analysis for healthcare professionals, especially radiologists.
Himashi Peiris, a Ph.D. candidate from the Faculty of Engineering, explains that their design introduces a “dual-view” AI system. One part of the system imitates how radiologists label medical images, while the other part evaluates the quality of AI-generated labels by comparing them to the limited labeled scans available.
Traditionally, medical experts annotate scans manually, which is a subjective and time-consuming process with potential errors and long waiting times for patients. The availability of large-scale annotated medical image datasets is limited due to the significant effort and expertise required for manual annotation.
The algorithm developed by Monash researchers enables multiple AI models to leverage the strengths of both labeled and unlabeled data. By learning from each other’s predictions, the algorithm enhances overall accuracy and outperforms state-of-the-art approaches, even with limited annotations.
The promising results of semi-supervised learning enable AI models to make more informed decisions, validate initial assessments, and provide more accurate diagnoses and treatment decisions. This research opens new possibilities for the medical imaging field.
The team’s next steps involve expanding the application to various types of medical images and developing an end-to-end product that radiologists can utilize in their practices.
Leading the study are Associate Professor Mehrtash Harandi and Ph.D. candidate Himashi Peiris from Monash University’s Faculty of Engineering, along with Associate Professor Zhaolin Chen, Dr. Munawar Hayat, and Professor Gary Egan from Monash Biomedical Imaging and the Faculty of Information Technology.
