New forms of cancer cell identification coupled with faster detection and better accuracy may enhance diagnostic capabilities. The purpose of this study is to improve recognition of minimal residual disease from peripheral blood samples. Spectral images are generated by optical microscopy using filtered broadband visible light elastically scattered from human blood and cancer cells. Exogenous tags, like CD markers may introduce a label bias and are not required. A training cell may be validated without detailed knowledge of intra-cellular spectra used to classify random cells. Spectral object classification is scalable to any number of cell types. Small samples of erythrocytes, leukocytes, Jurkat cancer and non-small lung cell adenocarcinoma are accurately classified and associated with unique spatial-spectral characteristics.