Key points from article :
A study led by researchers at Karolinska Institutet in Sweden explored how AI could outperform human experts in detecting ovarian cancer. Ovarian tumors are common but often detected by chance, and there is a shortage of ultrasound experts worldwide, which can delay diagnoses. The researchers developed neural network models to distinguish between benign and malignant ovarian lesions, testing them on over 17,000 ultrasound images from 3,652 patients across 20 hospitals in eight countries. The results showed that AI achieved an accuracy rate of 86.3%, surpassing both expert (82.6%) and non-expert (77.7%) examiners in identifying cancer.
These AI models could reduce unnecessary expert referrals and misdiagnoses, as demonstrated in a trial where AI support lowered referrals by 63% and the misdiagnosis rate by 18%. The study emphasizes that AI could be especially useful in places with fewer ultrasound specialists, providing valuable support in diagnosing hard-to-detect cancers. However, the researchers noted that further studies are necessary to fully understand the AI's potential and limitations in clinical settings. The team is currently conducting clinical studies to assess the safety and effectiveness of the AI tool at Södersjukhuset, with future research focusing on its impact on patient management and healthcare costs.
This study, published in Nature Medicine, highlights AI's role in healthcare, particularly in optimising resources and relieving pressure on medical experts. The team cautions that AI-based tools must be adaptable to various clinical environments and patient groups to be fully effective.
