Researchers from the Hong Kong University of Science and Technology (HKUST) have introduced a cutting-edge artificial intelligence model, Mixture of Modality Experts (MOME), designed to enhance non-invasive breast cancer diagnosis. Trained on China’s largest multiparametric MRI (mpMRI) breast cancer dataset, MOME delivers expert-level accuracy in classifying tumor malignancy—on par with radiologists possessing over five years of clinical experience.
Extensive Clinical Validation Underway
To ensure real-world effectiveness, MOME is currently undergoing large-scale clinical validation across more than ten hospitals and medical institutions. Participating sites include Shenzhen People’s Hospital, Guangzhou First Municipal People’s Hospital, and Yunnan Cancer Center. The findings have been published in the journal Nature Communications.
Addressing Diagnostic Complexity with Advanced AI
Breast cancer remains one of the most prevalent and life-threatening cancers affecting women globally. Accurate early detection, molecular subtyping, and treatment response prediction are critical to improving outcomes. Multiparametric MRI offers rich diagnostic information. However, integrating its multiple imaging sequences poses challenges, especially when some sequences are missing—a common issue in clinical settings.
To tackle this, the HKUST-led team collaborated with various medical institutions to build the largest mpMRI breast cancer dataset in China. They then developed MOME using a “mixture-of-experts” framework and a transformer-based architecture. This enables the model to robustly interpret multimodal data, even with incomplete imaging sequences.
A Leap Toward Personalized Cancer Care
MOME doesn’t just diagnose; it also supports molecular subtyping and predicts how patients will respond to treatment, including neoadjuvant chemotherapy, which is given before surgery to shrink tumors. Notably, the model also shows promise in identifying and subtyping triple-negative breast cancer, an aggressive and treatment-resistant form of the disease.
In trials, MOME demonstrated its capability to reduce unnecessary biopsies. It does this by accurately identifying benign cases among BI-RADS 4 patients, who typically present with suspicious imaging findings and moderate cancer risk. This improvement could significantly streamline the diagnostic process and reduce patient anxiety and clinical burden.
Interpretability and Integration into Clinical Practice
According to Prof. Chen Hao, Assistant Professor at HKUST’s Departments of Computer Science and Engineering, Chemical and Biological Engineering, and Life Science, MOME’s high adaptability and decision interpretability make it ideal for clinical integration. “MOME enhances diagnostic reliability and transparency,” he said, “underscoring the transformative power of AI in modern medical imaging.”
He added, “With rapid advancements in large AI models and imaging technologies, solutions like MOME will increasingly support clinicians. This will help deliver care that is more precise and personalized.”
Global Collaboration Behind the Innovation
As reported by medicalxpress, the study, titled “A Large Model for Non-Invasive and Personalized Management of Breast Cancer from Multiparametric MRI,” was a collaborative effort involving HKUST’s Smart Lab, Harvard University, Shenzhen People’s Hospital, PLA General Hospital, and Yunnan Cancer Center. Dr. Luo Luyang, a former postdoctoral fellow at HKUST and current researcher at Harvard, served as the study’s first author.
HKUST is paving the way for a new era in non-invasive, AI-powered breast cancer diagnostics through MOME. This innovation offers hope for earlier, more accurate, and more personalized patient management.