Feed

Breast cancer surgery urgently requires a rapid and objective intraoperative diagnostic strategy, as conventional frozen-section pathology depends on time-consuming staining and subjective interpretation, limiting timely surgical decision-making. However, existing label-free microscopic imaging approaches exclude conventional pathological information and lack diagnostic power. To overcome these limitations, a multimodal fusion framework based on unstained multi-angle orthogonal polarization micro-imaging (OPMI) and bright-field micro-imaging (BFMI) was proposed for rapid intraoperative breast cancer diagnosis. By superimposing and differencing multi-angle polarized images, optical contrast of anisotropic collagen fibers was markedly enhanced without staining, enabling objective visualization of collagen morphology and spatial alterations associated with malignancy. Local radial alignment and increased density of collagen fibers corresponding to invasive progression were identified. A dual-encoder fusion network integrating global and local feature representation improved lesion structure and boundary delineation. The method achieved 91.83% accuracy and area under the curve (AUC) of 0.9295, outperforming unimodal and bimodal approaches, offering a promising label-free intraoperative decision-support strategy.