A Novel Method Based on Watershed and Transfer Learning for Recognizing Immature Precursor Cells

It is an effective way to use digital image processing method to assist medical research. This paper presents a hybrid approach based on watershed and transfers learning method to automatically segment, characterize, and classify the particular immature precursor (IP) cells in bone marrow pathological (BMP) images. In segmentation phase, we use adaptive morphological reconstruction to accentuate the cell shapes and use improved marker-controlled watershed to segment the cells. Eleven morphological and statistical features are then extracted from those samples. In classification phase, we use transfer learning method to make use of the assistant sample set and generate a strong SVM classifier. Experimental results show the proposed method has a better performance, and the result lays the foundation for study of the correlations between the IP cells in BMP images and the relapse of acute myeloid leukemia (AML).