Toward a robust lesion detection model in breast DCE-MRI: adapting foundation models to high-risk women

2025-09-02

AI Summary

To address the low diagnostic accuracy and poor interpretability in differentiating benign from malignant lesions in dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) for high-risk breast cancer screening, this paper proposes an MST+KAN fusion framework. It employs a Medical Slice Transformer (MST), pre-trained on DINOv2, to extract discriminative features from individual MRI slices, and integrates a Kolmogorov–Arnold Network (KAN) with B-spline activation functions for nonlinear classification. An embedded attention mechanism further generates interpretable lesion-level heatmaps. This design jointly enhances model transparency and robustness against imbalanced and heterogeneous clinical data. Experimental results on an independent test set demonstrate that MST+KAN achieves an AUC of 0.80 ± 0.02—surpassing state-of-the-art baselines—while maintaining strong diagnostic performance and clinically meaningful interpretability.

Technology Category

• Computer Vision: Interpretability, Explainability, and Transparency
• Machine Learning: Transparent, Interpretable, Explainable ML
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Accurate breast MRI lesion detection is critical for early cancer diagnosis, especially in high-risk populations. We present a classification pipeline that adapts a pretrained foundation model, the Medical Slice Transformer (MST), for breast lesion classification using dynamic contrast-enhanced MRI (DCE-MRI). Leveraging DINOv2-based self-supervised pretraining, MST generates robust per-slice feature embeddings, which are then used to train a Kolmogorov--Arnold Network (KAN) classifier. The KAN provides a flexible and interpretable alternative to conventional convolutional networks by enabling localized nonlinear transformations via adaptive B-spline activations. This enhances the model's ability to differentiate benign from malignant lesions in imbalanced and heterogeneous clinical datasets. Experimental results demonstrate that the MST+KAN pipeline outperforms the baseline MST classifier, achieving AUC = 0.80 pm 0.02 while preserving interpretability through attention-based heatmaps. Our findings highlight the effectiveness of combining foundation model embeddings with advanced classification strategies for building robust and generalizable breast MRI analysis tools.

Problem

• Adapting foundation models for breast lesion detection
• Differentiating benign from malignant lesions in MRI
• Enhancing interpretability in clinical breast cancer diagnosis

Innovation

• Adapts Medical Slice Transformer foundation model
• Uses DINOv2 self-supervised pretraining for embeddings
• Employs Kolmogorov-Arnold Network with B-spline activations