The paper proposes a new framework, MorphoITH, which uses histopathologic slides to deconvolve intra-tumor heterogeneity (ITH) through identification of mutation states, molecular subtypes, and diagnostically distinct areas. Representing a critical challenge in precision oncology, ITH is driven by cancer cell evolution and contributes to tumor aggressiveness. The method aims to construct a self-supervised deep learning model that can accurately recognize morphological difference, quantify phenotypic diversity, and control for spurious sources of variation. The model was then prototyped on clear cell renal cell carcinoma (ccRCC) to evaluate the effect of key driver gene mutations in ccRCC and capture clinically-relevant features and morphological patterns. To investigate the role of morphological ITH in tumor evolution, multi-region whole-exome sequencing data was obtained from three patients (26 samples total), with their phylogenetic trees with the morphological similarity assessed by the framework. It was found that the morphological heterogeneity predicted by MorphoITH aligned with genetic evolutionary patterns, while observing exceptions that may reflect factors such as convergent evolution. This study presents MorphoITH as a promising framework for studying intra-tumor heterogeneity from histopathologic slides, as well as demonstrating potential to improve tumor sampling and advance precision oncology.