Automated image processing can be integrated with molecular profiling to provide a fuller portrait of cancer.
Pathologists routinely use visual examinations of cell types in tumour biopsies to direct patient care. However, it is hard to integrate these visual analyses with data from gene-expression studies.
Florian Markowetz and Yinyin Yuan at the University of Cambridge, UK, and their team came up with software that can analyse images of stained tissue sections to determine the identity and arrangement of cells in tumours. For some cell types, certain spatial patterns were associated with longer patient survival. However, an algorithm that combined image-based and gene-expression data predicted survival more accurately than algorithms that used either type of information alone.