Please use this identifier to cite or link to this item: http://hdl.handle.net/10603/187487
Title: Automatic feature extraction and classification of cell images for Cytopathology
Researcher: Gopakumar. G
Guide(s): Gorthi R K Sai Subrahmanyam
Keywords: cell segmentation; cell image; Boltzmann machine; convolutional neural network
cytopathology; IFC; microscopy; malaria; leukaemia
University: Indian Institute of Space Science and Technology
Completed Date: 2017
Abstract: Cytopathology is the analysis at cellular level for disease diagnosis. Every cell has standard morphology and typical count in unit volume constituting the cell signature. Depending on the pathological state of the individual, the signature may change and is the subject of cytopathology. Manual microscopic examination is the gold standard for cytopathology but is a tedious, skill demanding job and suffers from low throughput. Automated microscopy and more recently imaging flow cytometry (IFC) emerged to overcome these diand#64257;iculties and to standardise the result. However these systems used extensive robotic handling and/ or expensive and#64258;uid handling mechanisms, making them bulky, expensive and not suitable for resource limited clinics. In our research, we strive for developing very costeffective point-of care diagnostics platforms by using off-the shelf, low-cost components. However the low-cost instrumentation has introduced great challenges in processing the acquired data such as dealing with the focus shift, unlabeled,unstained data and imaging artefacts. We have overcome these challenges by designing, developing and employing sophisticated image analysis and advanced machine learning algorithms. We have proposed processing frameworks for both microscopy and IFC: a framework to automate malaria diagnosis in microscopy and a general framework for processing and classiand#64257;cation of cells in IFC. The frameworks include feasible preprocessing, novel cell segmentations, feature extraction as well as classiand#64257;cation. We have explored both the possibility of using conventional classiand#64257;ers (like support vector machine and nearest neighbour) and trending deep learning based classiand#64257;ers (based on restricted Boltzmann machine and convolutional neural network) and proposed classiand#64257;cation techniques even when the availability of labeled data for training is limited. The feasibility of the IFC framework is established by classifying leukaemia cell-lines (K562, MOLT, and HL60)
Pagination: xviii, 172p.
URI: http://hdl.handle.net/10603/187487
Appears in Departments:Department of Earth & Space Sciences

Files in This Item:
File Description SizeFormat 
01_title.pdfAttached File262.42 kBAdobe PDFView/Open
02_certificate.pdf37.95 kBAdobe PDFView/Open
03_declaration.pdf37.64 kBAdobe PDFView/Open
04_acknowledgements.pdf31.55 kBAdobe PDFView/Open
05_abstract.pdf29.12 kBAdobe PDFView/Open
06_table of contents.pdf64.82 kBAdobe PDFView/Open
07_list_of_tables.pdf51.92 kBAdobe PDFView/Open
08_list_of_figures.pdf104.35 kBAdobe PDFView/Open
09_abbreviations.pdf26.35 kBAdobe PDFView/Open
10_chapter 1.pdf314.68 kBAdobe PDFView/Open
11_chapter 2.pdf4.4 MBAdobe PDFView/Open
12_chapter 3.pdf4.36 MBAdobe PDFView/Open
13_chapter 4.pdf1.1 MBAdobe PDFView/Open
14_chapter 5.pdf1.43 MBAdobe PDFView/Open
15_chapter 6.pdf850.86 kBAdobe PDFView/Open
16_chapter 7.pdf60.64 kBAdobe PDFView/Open
17_bibliography.pdf143.27 kBAdobe PDFView/Open
18_appendix.pdf344.56 kBAdobe PDFView/Open
19_list of publications based on the thesis.pdf76.32 kBAdobe PDFView/Open


Items in Shodhganga are protected by copyright, with all rights reserved, unless otherwise indicated.