Robust Cell Image Analysis Techniques

Abstract

White blood cells (leukocytes) play an important role in the diagnosis of different diseases. The automatic detection of white blood cells is an unsolved issue in the field of medical imaging. Researchers from the fields of medicine and computer vision are engaged towards the analysis of white blood cell images. The objective of this work is to detect overlapped and single white blood cells accurately in order to use them for further analysis, such as white blood cell classification and disease detection. Due to the rapid development in the technology, wide range of techniques have been proposed for white blood cell counting, segmentation and classification, yet there is a need for automatic, accurate and robust white blood cell detection technique. In this thesis, an automatic white blood cell detection technique is proposed. The proposed technique utilizes the concept of iterative Otsu segmentation, marker controlled watershed transform, edge map and parametric circle approximation. The proposed technique is capableof detecting both separated and overlapped white blood cells in blood smear images. The extracted white blood cells are classified further into five types i.e. monocytes, basophils, neutrophils, lymphocytes and eosinophils. The cancerous and non-cancerous lymphocytic white blood cells are classified to detect acute lymphoblastic leukemia disease. The proposed technique is also tested to classify tumor and non-tumor MRI scan images of brain. The visual and quantitative comparisons (with state of art existing techniques) are performed to verify the significance of proposed techniques. The simulation results reveal that the proposed techniques are more accurate and robust as compared to state of art detection techniques.