BC-MRI-SEG: A Breast Cancer MRI Tumor Segmentation Benchmark
Anthony Bilic, Chen Chen
TL;DR
BC-MRI-SEG addresses the lack of public benchmarks for breast cancer MRI tumor segmentation by introducing a unified, multi-dataset benchmark. It uses ISPY1 and BreastDM for supervised training and RIDER and DUKE for zero-shot evaluation, comparing several 2D/3D/1D U-Net variants along with transformer- and adapter-based approaches. The authors provide preprocessing scripts, compile a comprehensive list of public breast MRI datasets, and release code for reproducibility. The results show that lightweight, context-rich 2D/1D variants (e.g., U-Net2.1D) can match or surpass more complex 3D models in both supervised and zero-shot settings, highlighting the importance of generalization across datasets. This benchmark has practical impact by enabling fair comparisons and guiding the development of robust models for clinical breast MRI segmentation.
Abstract
Binary breast cancer tumor segmentation with Magnetic Resonance Imaging (MRI) data is typically trained and evaluated on private medical data, which makes comparing deep learning approaches difficult. We propose a benchmark (BC-MRI-SEG) for binary breast cancer tumor segmentation based on publicly available MRI datasets. The benchmark consists of four datasets in total, where two datasets are used for supervised training and evaluation, and two are used for zero-shot evaluation. Additionally we compare state-of-the-art (SOTA) approaches on our benchmark and provide an exhaustive list of available public breast cancer MRI datasets. The source code has been made available at https://irulenot.github.io/BC_MRI_SEG_Benchmark.