A weakly supervised NMF method to decipher molecular subtype-related dynamic patterns in breast DCE-MR images

Area of Science:

• Biomedical Imaging
• Radiology
• Computational Biology

Background:

• Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is crucial for breast cancer diagnosis.
• Intratumoral heterogeneity presents a significant challenge in interpreting DCE-MRI data.
• Existing decomposition methods lack clinical relevance by operating in an unsupervised manner.

Purpose of the Study:

• To develop a weakly supervised method for DCE-MRI analysis that incorporates molecular subtype information.
• To improve the identification of intratumoral heterogeneity related to breast cancer subtypes.
• To enhance the diagnostic accuracy of breast cancer by analyzing pixel kinetics.

Main Methods:

• Proposed a weakly supervised nonnegative matrix factorization (WSNMF) method for DCE-MRI.
• WSNMF utilizes image-level subtype labels and discriminant nonnegative matrix factorization (NMF).
• Defined between- and within-class scatters on mean component coefficients for coarse-grained subtype information.

Main Results:

• WSNMF demonstrated superior performance in distinguishing Luminal A tumors from other subtypes.
• Achieved a higher area under the receiver operating characteristic curve (AUC) of 0.822 compared to other methods.
• Outperformed partitioning-based (KPC, TTP) and unsupervised decomposition (PCA, NMF) methods in classification accuracy.

Conclusions:

• The proposed WSNMF method offers a novel approach to DCE-MRI heterogeneity analysis.
• Leveraging intrinsic tumor characteristics improves breast cancer diagnosis.
• This weakly supervised method provides a valuable tool for understanding DCE-MRI patterns and molecular subtypes.