Real-time prediction of breast cancer sites using deformation-aware graph neural network

TL;DR

This paper presents a real-time deformation-predicting model for breast cancer sites by training a graph neural network on subject-specific FE simulations derived from MRI data. By encoding FE-derived tissue displacements into a surface-to-volume graph with distance-based and structured edges, the model accurately predicts tumor displacement during biopsy while achieving real-time inference. Key findings include sub-millimeter cancer localization accuracy (cancer RMSE ≈ 0.278 mm) and a DSC of ~0.977 on patient data, with speed-ups exceeding four thousand-fold compared to conventional FE. The approach offers a practical pathway to integrating physics-informed GNNs into indirect MRI-guided biopsy workflows, potentially improving precision and efficiency in breast cancer diagnosis and guiding broader needle-based interventions.

Abstract

Early diagnosis of breast cancer is crucial, enabling the establishment of appropriate treatment plans and markedly enhancing patient prognosis. While direct magnetic resonance imaging-guided biopsy demonstrates promising performance in detecting cancer lesions, its practical application is limited by prolonged procedure times and high costs. To overcome these issues, an indirect MRI-guided biopsy that allows the procedure to be performed outside of the MRI room has been proposed, but it still faces challenges in creating an accurate real-time deformable breast model. In our study, we tackled this issue by developing a graph neural network (GNN)-based model capable of accurately predicting deformed breast cancer sites in real time during biopsy procedures. An individual-specific finite element (FE) model was developed by incorporating magnetic resonance (MR) image-derived structural information of the breast and tumor to simulate deformation behaviors. A GNN model was then employed, designed to process surface displacement and distance-based graph data, enabling accurate prediction of overall tissue displacement, including the deformation of the tumor region. The model was validated using phantom and real patient datasets, achieving an accuracy within 0.2 millimeters (mm) for cancer node displacement (RMSE) and a dice similarity coefficient (DSC) of 0.977 for spatial overlap with actual cancerous regions. Additionally, the model enabled real-time inference and achieved a speed-up of over 4,000 times in computational cost compared to conventional FE simulations. The proposed deformation-aware GNN model offers a promising solution for real-time tumor displacement prediction in breast biopsy, with high accuracy and real-time capability. Its integration with clinical procedures could significantly enhance the precision and efficiency of breast cancer diagnosis.

Figures (5)

• Figure 1: Example of a finite element model constructed from an MR image. (a) Triplanar view of a breast MR image. The purple region indicates the cancerous area. (b) 3D surface data extracted from the MR image. (c) Finite element mesh representation.
• Figure 2: Graph construction process: (a) Graph nodes are created to match the nodes of the FE model. (b) Distance-based edge construction ($\mathcal{E}_d$): Nodes within 0.003 m of each other are connected to capture local spatial relationships. (c) Structured edge augmentation ($\mathcal{E}_s$): A subset of surface nodes and cancer surface nodes are directly connected to enhance long-range information propagation.
• Figure 3: Multi-layer GNN-based architecture used in this study. The input surface data is first passed through a linear layer with ReLU activation, followed by 8 GNN layers (each with ReLU). The model output is then generated via a 2-layer MLP, effectively reducing dimensionality.
• Figure 4: Exemplar deformed configuration of cancerous region in three patient datasets ('H1', 'H2', and 'H3'). For each dataset, the ground truth (blue) and GNN-predicted (red) deformed shapes are visualized in 3D, as well as in three 2D planes: axial (xy), coronal (yz), and sagittal (xz).
• Figure 5: Boxplots of global RMSE, cancer RMSE, and DSC for three real patient datasets ('H1', 'H2', and 'H3'). The black horizontal line inside each box represents the median, while the box spans the interquartile range (IQR), from the first quartile (Q1) to the third quartile (Q3). The whiskers extend to the lower (Q1 – 1.5×IQR) and upper (Q3 + 1.5×IQR) bounds, indicating the range of non-outlier data points. Outliers beyond the whiskers are shown as individual points.