Deep Learning-Based Analysis of Microscopic Hyperspectral Images of HER2-Stained Breast Cancer Sections for Molecular Status Prediction.

Authors

Lin Yan, Yan Wan, Haonan Liu, Wenlou Liu, Yang Zhao, Hongmei Wang, Juanjuan Tang, Xuanyu Chen, Zhengxiang Han, Xiaobing Qin

Published in

Journal of biophotonics. Volume 19. Issue 11.

Abstract

HER2 assessment in breast cancer relies on immunohistochemistry (IHC) with reflex in situ hybridization for equivocal cases, while conventional bright-field imaging provides limited quantitative optical information. We investigated microscopic hyperspectral imaging (HSI) of HER2-stained sections for slide-level molecular status prediction. HSI data from 43 cases (21 HER2-positive, 22 HER2-negative; 400-1000 nm, 128 bands) were acquired between January 2024 and August 2025. After calibration, optical-density conversion, tissue masking, and Savitzky-Golay smoothing, we developed SpectralMix-MIL, a weakly supervised multiple-instance learning model combining learnable spectral projection (K = 32) with attention pooling. On a patient-level 7:3 train/test split, SpectralMix-MIL achieved an AUROC of 0.923 and AUPRC of 0.910. Repeated stratified fivefold cross-validation yielded mean AUROC 0.904 ± 0.041 and AUPRC 0.892 ± 0.045, while leave-one-out cross-validation yielded AUROC 0.897. These findings support microscopic HSI for quantitative breast cancer molecular phenotyping, pending larger external validation.

PMID:
42995139
Bibliographic data and abstract were imported from PubMed on 07 Nov 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.