Single-cell RNA sequencing is becoming an increasingly common tool to investigate the cellular population and patients' outcomes in cancer research. However, due to the sparse data and the complex tumor microenvironment, it is challenging to identify neoplastic cells that play important roles in tumor growth and disease progression. This challenge is exaggerated in the research of blood cancer patients, from whom the neoplastic cells can be highly similar to normal cells. Here, we present partCNV/partCNVH, a statistical framework for rapid and accurate detection of aneuploid cells with local copy number deletion or amplification. Our method uses an expectation–maximization (EM) algorithm with mixtures of Poisson distributions while incorporating cytogenetic information to guide the classification (partCNV). When applicable, we further improve the accuracy by integrating a hidden Markov model for feature selection (partCNVH). We evaluate the performance of the proposed methods using extensive simulation studies and three scRNA-seq datasets from patients with blood cancers. In these applications, our proposed methods demonstrate favorable accuracy and more interpretable results compared with existing methods. In the real data analysis, our results identify many biological processes that have or have not been previously validated, implying multiple biological processes involved in the oncogenesis of myelodysplastic syndromes and acute myeloid leukemia.