Digital staining of multispectral microscopic images. Application to breast cancer indentification

Abstract:

The purpose of this work is the classification of non-stained breast tissue to discriminate between tumour and non-tumour samples. This is achieved using deep learning techniques and multispectral imaging (MSI), acquired with different wavelengths, from 425 nm to 700 nm and a spectral step of 4 nm. An analysis was performed to determine whether the different spectral bands to provide additional information with respect to images acquired with ordinary cameras. As a result, it has been identified the classification model implemented with MSI that obtains the closest hit rate to the stained sample image classification model. In addition, several digital staining methods have been implemented by applying deep learning techniques. Virtual staining would be performed after classification to corroborate the results, allowing to advance in the analysis of structural interactions at the cellular and tissue levels.