We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

eess.IV

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Electrical Engineering and Systems Science > Image and Video Processing

Title: All You Need is Color: Image based Spatial Gene Expression Prediction using Neural Stain Learning

Abstract: "Is it possible to predict expression levels of different genes at a given spatial location in the routine histology image of a tumor section by modeling its stain absorption characteristics?" In this work, we propose a "stain-aware" machine learning approach for prediction of spatial transcriptomic gene expression profiles using digital pathology image of a routine Hematoxylin & Eosin (H&E) histology section. Unlike recent deep learning methods which are used for gene expression prediction, our proposed approach termed Neural Stain Learning (NSL) explicitly models the association of stain absorption characteristics of the tissue with gene expression patterns in spatial transcriptomics by learning a problem-specific stain deconvolution matrix in an end-to-end manner. The proposed method with only 11 trainable weight parameters outperforms both classical regression models with cellular composition and morphological features as well as deep learning methods. We have found that the gene expression predictions from the proposed approach show higher correlations with true expression values obtained through sequencing for a larger set of genes in comparison to other approaches.
Comments: 14 pages, 4 figures, 1 table
Subjects: Image and Video Processing (eess.IV); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Quantitative Methods (q-bio.QM)
Cite as: arXiv:2108.10446 [eess.IV]
  (or arXiv:2108.10446v2 [eess.IV] for this version)

Submission history

From: Muhammad Dawood [view email]
[v1] Mon, 23 Aug 2021 23:43:38 GMT (763kb)
[v2] Thu, 26 Aug 2021 10:45:21 GMT (844kb)

Link back to: arXiv, form interface, contact.