The histopathological diagnosis in malignances requests well trained specialists and multi-step operational procedures for sample preparation. Faster and more objective evaluation protocols should be implemented to give support to the pathologists. The Quantitative Phase Imaging based methods are biological-proved to be efficient in revealing important characteristics of the living structures without any labeling. These can be further exploited for an automatic evaluation of complex tissues. Using an off-axis Digital Holographic Microscopy setup, biopsies of two histological origins: cerebral (grade II glioma and grade IV glioblastoma) and colonic malignancies (dysplastic and malignant colonic adenomatous polyps), were investigated. Various parameters of quantitative phase shift maps (QPMs) were computed (mean, variance, median, kurtosis, skewness, energy, entropy). The possibility of automatic discrimination of tumor tissues having different structural complexity and presenting various malignancy grades was evaluated using supervised machine learning algorithms. The analysis of phase shift maps has successfully discriminated between levels of malignancy with high statistical confidence in the case of gliomas. Moreover an algorithm with the ability to classify the tissue biopsies in different malignant stages using parameters based on QPMs has been implemented on glioma tissues having a high level of homogeneity. In case of colonic polyps, the heterogeneity of the multilayered tissue demanded QPMs analysis to be performed on selected area of interest even though some statistical differences were obtained for global evaluation of phase shift distributions. In case of colonic polyps, for a good accuracy of classification algorithm a larger library of QPMs is under construction.