Breakthrough AI Model Achieves Expert-Level Accuracy in Non-Invasive Breast Cancer Diagnosis with MRI

Researchers from the Hong Kong University of Science and Technology (HKUST) have introduced a cutting-edge artificial intelligence (AI) model called Mixture of Modality Experts (MOME) that significantly advances the non-invasive diagnosis of breast cancer using MRI data. Trained on China’s largest multiparametric MRI (mpMRI) dataset for breast imaging, MOME demonstrates diagnostic accuracy comparable to experienced radiologists with over five years of expertise. This innovative model not only classifies tumor malignancy accurately but also supports molecular subtyping and predicts response to neoadjuvant chemotherapy, enabling more personalized treatment planning.

MOME employs a sophisticated 'mixture-of-experts' framework combined with transformer architecture, allowing it to effectively fuse information from multiple imaging modalities. Crucially, it remains robust even when some MRI sequences are missing, overcoming a common challenge in clinical imaging where comprehensive data is often unavailable. Currently undergoing extensive validation across over ten hospitals—including Shenzhen People’s Hospital, Guangzhou First Municipal Hospital, and Yunnan Cancer Center—the model aims to prove its effectiveness in real-world clinical settings.

Breast cancer remains among the most prevalent and deadly cancers worldwide, emphasizing the need for early detection and precise subtyping. Traditional mpMRI provides a wealth of diagnostic information but integrating its different sequences can be challenging. The MOME model addresses these issues by leveraging a vast dataset and advanced AI techniques to improve diagnostic reliability and support non-invasive tumor classification.

In trials, MOME excelled not only in accuracy but also in reducing unnecessary biopsies, identifying benign cases among patients with suspicious findings (BI-RADS 4), and predicting responses to treatments like chemotherapy, especially in aggressive subtypes such as triple-negative breast cancer. Professor Chen Hao of HKUST highlighted the transformative potential of the system, stating that it enhances decision transparency and could seamlessly integrate into clinical workflows. He expressed optimism that such models will increasingly empower clinicians and enhance patient outcomes.

The research, published in Nature Communications, signifies a major step toward AI-driven personalized cancer care, with collaborative efforts spanning HKUST, Harvard University, and prominent Chinese hospitals. Leading the research was Dr. Luo Luyang, former HKUST postdoc and current Harvard researcher. The study underscores the significant promise of large AI models and advanced imaging technologies in transforming breast cancer diagnosis and management.

For more details, see the publication: Nature Communications. Source: Medical Xpress