Use this URL to cite or link to this record in EThOS:
Title: Computer vision systems for zebrafish larvae analysis
Author: Alsaaidah, Bayan Abd AlKareem
ISNI:       0000 0004 7970 5176
Awarding Body: University of Liverpool
Current Institution: University of Liverpool
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
Biologists and pharmacologists commonly use zebrafish embryos during the testing of drugs due to their properties and high genetic similarity with mammals. The testing of these substances is a tedious and painstaking process, carried out manually by trained experts who determine whether the embryos have been deformed or killed as a result of administering the chemical. Computer vision presents efficient solutions for such aquaculture research problems by the application of machine learning systems. However, having accurate and cost-effective monitoring and analysis systems still has challenges due to the small size and the high similarity of the samples and the unwanted position of the samples due to their free swimming. In this thesis, a set of novel and cost-effective systems are presented consisting of detection, classification, identification, and counting zebrafish samples. In the first part, two applications are proposed to address several problems that the biologists face in their work. First one is a novel and accurate system for detecting and counting the number of transparent live eggs inside a petri dish that may contain hundreds of eggs either dead or live with other types of objects. The second one is counting the adult fish inside a tank that confuses the biologists in their work without transferring the fish from tank to another as an attempt for counting them. In the second part, an image analysis pipeline is presented from the data acquisition process to data analysis and classification system to get a final result about the embryo status. The proposed work aims to identify the health and detect the abnormalities of zebrafish embryos using scanned images and based on computer vision algorithms. The images are comprised of many features which could be extracted automatically or manually. In the third part, a multi-label classification system for zebrafish embryo deformities based on microscopic images is presented. The proposed system aims to determine the numbers, health, and presence of abnormalities in zebrafish larvae using a classification system with high-throughput delivering results faster than the manual process, and assisting in the pharmacological and toxicological experiments. The novelty here is having a malformation classification system depending on different feature types and identifying all the deformity classes that may appear on a sample body at the same time using binary relevance multi-label algorithm. According to the proposed application purposes and based on computer vision algorithms, this thesis successfully addressed most problems and provided efficient models for analysing, classifying, and counting the zebrafish larvae. The proposed models provided the biologists by a novel and cost-effective computer vision models.
Supervisor: Alnuaimy, Waleed Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral