Identification of Myeloproliferative Neoplasms Using Deep Learning

Myeloproliferative Neoplasms (MPNs) are a heterogeneous group of disorders characterized by proliferation of one or more hematologic cell. These myeloproliferative disorders have different morphological features associated with it. Microscopic studies and morphological evaluation becomes mandatory in all these cases to reach a proper diagnosis. In this paper, we are trying to exploit the morphological features using deep learning techniques to narrow down the region of interest of MPNs. Here semantic segmentation is performed and the various types of MPNs (Benign, ET, MF, PV and CML) are classified. To perform this task, we have used MobileUNet and ResUNet++ deep learning network architectures and the performance is evaluated using F-scores and accuracy of the corresponding classes. The two models were compared and MobileUNet model is giving a better performance with an average F-score of 62% and ResUNet++ is having an average F-score of 59%.