Identification of nosema cells using microscopic images

Abstract

Research in the field of computer vision and intelligent systems has become increasingly vast and extensive to meet the needs and conditions of all users. Additionally, new machine learning architectures have shown profound results and made the interpretation and analysis of media more robust and efficient. The robustness and efficiency of these new architectures, coupled with technology development, have made a new area of application and opened the door for new research more beneficial for the end-user. Indeed, in the field of biology, microscopic image analysis has led to an important evolution in terms of the creation of new diagnostic support systems. The purpose of the latter is to provide practitioners with an automatic interpretation of microscopic images to allow an exploitation of the cells of such a studied disease. Different segmentation approaches have been proposed in the literature, but a method has yet to be deemed optimal for only a specific application. Therefore, it can be admitted that there is no universal method for segmentation; rather it depends on the type of knowledge sought. This thesis is articulated around the axis of segmentation methods, highlighting the crucial dependence on the specific type of knowledge being sought. The main objective of this work is to propose methods and algorithms to help recognize the cells of Nosema disease in microscopic images and make the diagnosis. These methods are very helpful in many fields and present an important pre-work for many applications. To achieve the objectives outlined in this thesis, various approaches such as: Machine Learning (ML), Deep Learning (DL, the newest and most efficient algorithm in machine learning techniques), and Augmentation Data (AD) are implemented and explored. As such, in this thesis, image processing tools will be used to calculate interesting features of Noema cells, and computer vision techniques, ML, DL, and AD techniques will be employed to recognize them. Finally, an automatic algorithm for cell identification and counting will be implemented. The automated system performs well in the diagnosis task, demonstrating high accuracy across four Nosema infection levels.