Hydrogel multimode fibers: enabling imaging and intelligent recognition of breast tumors.

Early-stage breast cancer lesions are often hidden within mammary ducts only a few hundred micrometers in diameter, where conventional silica-based fiber endoscopes suffer from severe mechanical mismatch with ultrasoft tissue. Hydrogel multimode optical fibers (HMMFs) offer excellent mechanical compliance and biocompatibility but are typically hindered by high optical loss and poor long-term stability due to dehydration. Here, we present a novel biocompatible HMMF for imaging and intelligent recognition of breast tumors. Micron-scale polyacrylamide (PAM) hydrogel fibers are continuously fabricated via a draw-spinning process. This process promotes the evaporation of free water while preserving tightly bound water within the polymer network, enabling stable optical guidance from the visible to the near-infrared region with an optical loss as low as 1.05 dB/cm at 980 nm. To decode the complex speckle patterns transmitted through the HMMFs, we develop a deep neural network termed BiMamba-UNet. This model achieves high-fidelity image reconstruction with average structural similarity (SSIM) values of 0.913 and 0.786 on MNIST and Fashion-MNIST datasets, respectively. Crucially, benefiting from the robust hydration retention of the material, the imaging system demonstrates negligible performance degradation over 7 days. Furthermore, as a clinical proof-of-concept, a two-stage framework is proposed to enable the accurate reconstruction and recognition of H&E-stained breast tumor sections. This work establishes a robust platform for flexible multimodal fiber imaging and opens new avenues toward in vivo optical biopsy.