Acute lymphoblastic leukemia (ALL) is a main health problem throughout the world. Therefore, fast and exact diagnosis is the most crucial factor for providing efficient management and treatment methods. The conventional diagnostic tools, based on the morphological and cytochemical investigation of blood and bone smears, are usually not specific and laborious. Thus, they often result in diagnostic errors and delay in treatment initiation. In this paper, ALL-diagnosing methods based on the convolutional autoencoder (CAE) was proposed to reduce the amount of data, and then convolutional neural network (CNN) was applied to identify ALL. The design method employed deep neural networks to recognize the features of the cells in question and then distinguish them as either leukemic or healthy cell types. The proposed laboratory method, with the use of the curated datasets of annotated pathological images of normal lymphoid progenitor cells, aimed to tackle the challenges related to the lack of curated datasets with annotated images of these cells. These challenges are believed to be linked to imprecise and time-consuming leukemia diagnosis and cure process. The simulated results confirmed the efficiency of the suggested technique, where CAE showed a correlation coefficient of 0.987 for lymphoblastic cells and CNN had an accuracy rate of 99.92% in ALL diagnosis. Such data demonstrated the capability of deep-based methodologies to fight leukemia.