Segmentation and counting of multiple myeloma cells using IEMD based deep neural network. (November 2022)

• Record Type: Journal Article
• Authors: Rasal, Tushar; Veerakumar, T.; Subudhi, Badri Narayan; Esakkirajan, S.
• Is Part Of: Leukemia research. Volume 122(2022)
• Publication Date: 2022-11
• Subjects: Multiple myeloma -- Deep learning -- Image segmentation -- Fluorescence microscopy; Leukemia -- Periodicals; Leukemia -- Periodicals; Leucémie -- Périodiques; Leukemia; Periodicals; Electronic journals; Electronic journals; 616.9941905
• DOI: 10.1016/j.leukres.2022.106950
• Languages: English
• ISSNs: 0145-2126

Abstract

Abstract: In biomedical image analysis, segmentation of cell nuclei from microscopic images is a highly challenging research problem. In the computer-assisted health care system, the segmented microscopic cells have been used by many biological researchers for the early prediction of various diseases. Multiple myeloma is one type of disease which is also term as a plasma cell cancer. The segmentation of the nucleus and cell is a very critical step for multiple myeloma detection. Here, In this work, we have designed two modules. One is for recognizing the nucleus of myeloma cells with a deep IEMD neural network, and the other is for differentiating the cell i.e cytoplasm. The different IMFs provides detailed frequency component of an image which are used for feature extraction. This will significantly improves the performance. We proposed a new counting algorithm for counting the myeloma-affected plasma cells in this paper. An algorithm for counting overgrowth plasma cells within the myeloid tissue has been developed using the Python TensorFlow framework. Experimental outcomes on SegPC datasets substantiate that, the proposed deep learning approach outperforms other competitive methods in myeloma recognition and detection. The result of this research indicates that, the proposed image segmentation mechanism can recognize multiple myeloma with superiority. Early detection of multiple myeloma at the initial stage increases the chances to cure patients. Highlights: Proposed two modules: for recognizing the nucleus of MM cells and for differentiating the cell membrane i.e cytoplasm. In-depth feature extraction with IEMD and IMFs which step will significantly improve the performance. Developed new counting algorithm with Python for counting overgrowth plasma cells within the myeloid tissue. Validation with publicly available SegPC datasets for various imaging modalities.