Optical microscopic examination of stained tissue by pathologists is the gold standard for the diagnosis of most cancers. Due to the human element involved, however, the process is slow, decisions are often complicated by subjective opinions and the uncertainty in diagnoses can affect therapy. Infrared spectroscopic imaging or hyperspectral molecular imaging, as opposed to optical wideband imaging, has been proposed as a viable alternative to provide automated, accurate, reproducible and useful diagnoses. Data processing to enable these applications, however, is not straightforward. Here we discuss recent advances in automatically profiling tissue and present the complexity and numerical strategies to address issues involved. Using breast cancer as an example, we show the importance of integrating statistical and mathematical tools into the analysis framework.