Histology images in spatial transcriptomics: a critical evaluation of their effectiveness in spatial domain detection

Spatial transcriptomics (ST) has become a powerful tool for studying gene expression at the spatial level, as the position of a given spot provides crucial information about its phenotype. One key application of downstream analysis on these ST datasets is identifying spatial domains within tissue regions. However, the accuracy of spatial domain detection can be hindered by technical noise and low resolution due to technological limitations. Some ST technologies now provide high-resolution histology images of the original tissue sample prior to sequencing, offering complementary information. Incorporating these histology images presents a promising approach to addressing these limitations. However, there is a scarcity of reviews on recent methods that use histology images for spatial domain detection tasks. Also, it remains unclear how much improvement is gained by these methods when integrating histology image with ST data over using ST data alone. This article presents an analysis of various methodologies that integrate histology images for spatial domain detection. We describe the specific challenges these methods address in incorporating histology images alongside gene expression data, highlighting both their commonalities and distinctions. Furthermore, we design a statistical test to systematically determine the significance of domain quality improvement with the additional modality of histology images, if any, by comparing the performance of methods run both with and without the histology data. Furthermore, we conduct statistical tests to determine the significance of histology images in improving spatial domain detection tasks. Our findings indicate no notable enhancement in spatial domain detection when histology images are used as an additional source of information alongside gene expression data for the methods under study. Building on these results, we investigate the possible reasons why histology images may fail to improve performance and discuss future directions for their effective integration in spatial domain detection. Source code is available at Github ( https://github.com/asifajrof/StHistDDComp ).