Multi-spectral virtual staining for label-free histopathology using deep neural networks

Virtual staining (VS) has gained a significant interest for label-free characterization of unstained tissue-sections. The concept of VS lies on translating a label-free imaging technique such as auto-fluorescence for generating an equivalent histologically stained image using deep-neural-networks. Different endogenous fluorophores present in the sample have distinct excitation and emission bands, which can provide stain-free specificity of molecules. However, conventional technique for studying emission of different molecules, restrict molecular specificity in the measurements. The present study explores multiple-spectral (MS) bands in label-free imaging to gain richer information about distribution and diversity of these fluorophores. Our study includes a comparative analysis using MS and single long-pass (SLP)-approach for VS of unstained tissue-sections. It is observed that MS-approach provides a better training input to the model enhancing discriminative power of the neural network, even when trained for lesser epochs. The Structural-Similarity-Index-Measure (SSIM) map shows that use of MS-approach for training the model provides a high-SSIM of 0.94 whereas, SLP-approach provides an SSIM-values of 0.62. The ROC area of a model quantifies the ratio of true-negative and false-positive outputs.The testing results show that MS-approach (ROC ≈ 0.98) outperforms the SLP-approach (ROC ≈ 0.78) even at a smaller training time. The versatility of MS-approach is tested by incorporating it for different tissue sections of multiple thicknesses providing consistently high value of ROC area (>0.92). The proposed approach offers a time-efficient and reagent-free add-on to well-established VS techniques that can be further extended for multiple biomedical imaging applications.

Authors

Institutions

Publication Details

Journal
Methods and Applications in Fluorescence
Published
2026-11-18
DOI
https://doi.org/10.1088/2050-6120/aec410
Primary Topic
AI in cancer detection
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Multi-spectral virtual staining for label-free histopathology using deep neural networks

Krishna Agarwal, Varun Surya, Himanshu Joshi, Deepika Mishra et al.
4 links
Methods and Applications in Fluorescence
AI in cancer detection
article

Multi-spectral virtual staining for label-free histopathology using deep neural networks

Krishna Agarwal, Varun Surya, Himanshu Joshi, Deepika Mishra, Dalip Singh Mehta, Akshay Chandrakant Trimukhe, Atul Kumar
article en

Abstract

Virtual staining (VS) has gained a significant interest for label-free characterization of unstained tissue-sections. The concept of VS lies on translating a label-free imaging technique such as auto-fluorescence for generating an equivalent histologically stained image using deep-neural-networks. Different endogenous fluorophores present in the sample have distinct excitation and emission bands, which can provide stain-free specificity of molecules. However, conventional technique for studying emission of different molecules, restrict molecular specificity in the measurements. The present study explores multiple-spectral (MS) bands in label-free imaging to gain richer information about distribution and diversity of these fluorophores. Our study includes a comparative analysis using MS and single long-pass (SLP)-approach for VS of unstained tissue-sections. It is observed that MS-approach provides a better training input to the model enhancing discriminative power of the neural network, even when trained for lesser epochs. The Structural-Similarity-Index-Measure (SSIM) map shows that use of MS-approach for training the model provides a high-SSIM of 0.94 whereas, SLP-approach provides an SSIM-values of 0.62. The ROC area of a model quantifies the ratio of true-negative and false-positive outputs.The testing results show that MS-approach (ROC ≈ 0.98) outperforms the SLP-approach (ROC ≈ 0.78) even at a smaller training time. The versatility of MS-approach is tested by incorporating it for different tissue sections of multiple thicknesses providing consistently high value of ROC area (>0.92). The proposed approach offers a time-efficient and reagent-free add-on to well-established VS techniques that can be further extended for multiple biomedical imaging applications.

Methods and Applications in Fluorescence
All India Institute of Medical Sciences (IN), Indian Institute of Technology Delhi (IN), UiT The Arctic University of Norway (NO)
Indian Council of Medical Research, Council of Scientific and Industrial Research, India
AI in cancer detection
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon