On the Study of Childhood Medulloblastoma Auto Cell Segmentation from Histopathological Tissue Samples.

Whole slide imaging in histopathology is one of the most important aspects of computational pathology. Nucleus identification and extraction can play a critical part in digital microscopic examination. This work is an extension of our previous published work on childhood medulloblastoma biopsy machine learning classification where the classifier was based on ground truth annotated data. However complete automation would entail automatic segmentation of the cells. The paper explores various segmentation techniques for cell identification from biopsy tissue samples of childhood medulloblastoma microscopic images based on conventional machine learning methods. The study is based on indigenous patient data collected from medical centers of the region. The performance of the segmentation algorithms was compared using Jaccard and Dice coefficient metric.