This article studies the use of deep learning in predicting genetic aberrations in acute myeloid leukemia (AML) from single-cell images. To conduct this research, the authors first compiled a dataset of more than 2,000,000 single-cell images from 408 patients with AML. Then, using this dataset, a convolutional neural network was trained for the prediction of various therapy-relevant genetic alterations. The main findings of this paper include significant predictions of presence of favorable risk genetic alterations according to the ELN 2017 classification by the deep neural network. In addition, in the temporal validation cohort the ELN favorable risk status was still predictable by the neural network with a median AUROC of 0.64. The authors also tested whether their neural network could directly predict CBFB::MYH11 from filtered single-cell images of eosinophils and they found that the patient-level AUROC for the predictions using 5- fold cross validation was 0.93. Lastly, the FLT3-ITD mutation status could be significantly predicted using the neural network with a patient-level AUROC of 0.79 and the proposed pipeline in the study predicted MRC cytogenetics with a patient-level AUROC of 0.65. Overall, this study introduces a pipeline that extracts genetic features from Pappenheim-stained whole bone marrow smear scans and uses 2 deep learning models to classify cell regions and predict genetic features. This pipeline was able to predict NPM1 mutations, FLT3 mutations, the ELN 2017 favorable genetic risk group, and CBFB::MYH11 rearrangements with high AUROCs.