Cancer, a worldwide health challenge and prominent cause of death, can be more effectively treated with early detection, thus driving the development of detection methods. Mid-IR has been used for label-free chemical imaging to detect cancerous tissues and cells [1,2], as an alternative to pathology approaches based on chemical staining. In collaboration with bio-medical researchers, we are using IR to study cancer cells, with the goal to distinguish between healthy, cancerous and metastasizing cells in tumor-on-a-chip systems. While FTIR excels in tissue sections [3], water-filled channels necessitate a mid-IR photothermal (MIP) approach. Our custom setup can image absorption through water, with the only adaptation being an IR transparent carrier. We report on our initial experiments using machine learning and IR spectroscopy for cell classification, comparing conventional FTIR microscopy to our MIP setup.