Background: The assessment of bone marrow (BM) cells is a central component for the diagnosis of AML. However, the manual assessment of BMS and counting of differentiated cells is a time-consuming method that depends on the experience and abilities of cytomorphologists and is not fully reproducible.Aims: In this study, we develop a pipeline for the automatic segmentation and classification of cells from BMS of patients with AML using Machine Learning (ML) and Computer Vision techniques.Methods: Whole BMS slides from 647 patients diagnosed with AML between 1/2017 and 12/2020 in Toulouse University Hospital were numerized using Panoramic 250 Flash II (3DHISTECH) scanner to obtain high-definition pictures in a.MRXS format, providing multiple magnification levels whose highest corresponds to 823 X (121nm/pixel). To analyze BMS images, we designed a pipeline separated into 3 steps as described below: first, we pre-selected tiles at the deepest zoom level most likely to contain exploitable cells to define areas of interest (AOI). For this purpose, we proposed a two-tier approach to filter fixed-size tiles on two zoom levels consecutively, using a combination of a pixel-wise k-means method and decision trees using textural features. Second, we extracted BM cells from the AOI. In this step, we considered two approaches: segmentation with an in-house segmentation software, which implements a genetic algorithm to find the cells, and a deep-learning-based model called Faster-RCNN, used for cell detection in the state-of-the-art. The former method only requires a low number of annotated images to train, but the training of the second one is much faster. In the third and final step, we combined the computation of radiomics features with ML methods to classify the extracted single cell images in a semi-supervised and unsupervised manner.Results: We were able to design a pipeline allowing automatic BM cell segmentation and classification, to identify between 1k and 1.5k cells from each BMS. At a lower magnification, we were also able to design a pipeline to identify megakaryocytes. We will show the entire multi-step pipeline and examples of automatically extracted and classified cells.Summary/Conclusion: Running our AI-based pipeline shows encouraging results in terms of extraction and classification of cells from BMS of AML. Our work will lead to the development of an AI tool to support cytomorphologists in the assessment of BMS from patients with AML. It will also permit the constitution of an AML bone marrow cells library