In most European countries, cancer diagnosis is based on the examination of haematoxylin–eosin (HE)–stained tissue sections by experienced pathologists. Although immunohistochemical stainings provide additional diagnostic information, they are not routinely applied in all cases due to their technical complexity, time requirements, and cost. This results in a medical need for fast, non-invasive, and label-free alternatives to conventional immunohistochemistry.

The STAIN-IT project addresses this need by developing a computational, label-free approach to immunohistochemistry based on multimodal optical imaging. Non-invasive techniques such as coherent anti-Stokes Raman scattering, second harmonic generation, and two-photon-excited fluorescence are used to acquire multimodal image data. These data are analysed using deep neural networks, in particular convolutional neural networks, to computationally approximate immunohistochemical stainings.

A central focus of STAIN-IT is the systematic analysis and quantification of the nonlinear behaviour of these networks. By applying suitable approximation and interpretation methods, the models are intended to move beyond black-box approaches and enable a quantitative description of tissue-specific changes.

STAIN-IT thus establishes, for the first time, the basis for fast, label-free, non-invasive, and labour-efficient computational immunohistochemical diagnostics that can be integrated into clinical routines. Potential applications include intraoperative analysis of the proliferation marker Ki-67 in frozen section diagnostics and the use of collagen IV as a quality marker in tissue-engineered products.