AI mimics the process of asking for a second opinion

Researchers from Monash University highlight that the system enables multiple AI models to learn from each other’s predictions

A clinician wearing a white surgical gown and a face mask looks closely at a medical scan pinned to the wall
Pixabay/Dmitriy Gutarev

New research has explored the potential of imitating the process of asking a second opinion within medical artificial intelligence (AI).

As part of a study published in Nature Machine Intelligence, scientists developed a new co-training AI algorithm that learns from the predictions of multiple AI models to improve accuracy.

PhD candidate Himashi Peiris, of Monash University, shared that there is competition between two components of a ‘dual view’ AI system.

“One part of the AI system tries to mimic how radiologists read medical images by labelling them, while the other part of the system judges the quality of the AI-generated labelled scans by benchmarking them against the limited labelled scans provided by radiologists,” she shared.

Peiris highlighted that the AI system has produced “groundbreaking results” in semi-supervised learning.

“It demonstrates remarkable performance even with limited annotations, unlike algorithms that rely on large volumes of annotated data,” she said.

“This enables AI models to make more informed decisions, validate their initial assessments, and uncover more accurate diagnoses and treatment decisions,” Peiris elaborated.

The next phase of the research will involve expanding the application of the AI system to work with different types of medical images, and developing a product that radiologists can use in practice.