E-Thesis 71 views 22 downloads
Applying a deep neural network based approach to automating the Micronucleus (MN) assay / QIELLOR HAXHIRAJ
Swansea University Author: QIELLOR HAXHIRAJ
PDF | E-Thesis – open access
Copyright: The author, Qiellor Haxhiraj, 2021.Download (5.7MB)
The Micronucleus (MN) Assay is a test mandated for use in genetic toxicology testing by regulatory bodies such as the Food and Drug administration (FDA). An increased quantity of MN is an indication of chromosomal damage which can be characterised into chromosomal breakage (caused by a clastogen) an...
|Degree level:||Master of Research|
No Tags, Be the first to tag this record!
The Micronucleus (MN) Assay is a test mandated for use in genetic toxicology testing by regulatory bodies such as the Food and Drug administration (FDA). An increased quantity of MN is an indication of chromosomal damage which can be characterised into chromosomal breakage (caused by a clastogen) and chromosomal loss (caused by an aneugen). By comparing a dose response, estimates can be made into the potency of the chemical. Historically the cell scoring procedure takes place through the ‘gold standard’ of manual scoring by light microscopy following staining. However, despite being classed the gold standard, this method is laborious and subjective, with archiving of results not a possibility. This leads to the need to develop a new technique to streamline the process, whilst still maintaining accuracy. The result is the creation of a ground truth based deep learning algorithm. By using imaging flow cytometry to carry out the MN assay, a ground truth was created, consisting of different cellular types, including MN. By scoring these images manually by eye, a ground truth of images to teach the deep-learning algorithm is created. By applying a deep neural network, the algorithm uses multiple layers to differentiate information, mimicking the way neurons work in the brain. This approach allows for differentiation between different cellular types based on the ground truth images scored. By assessing more images, the accuracy is further increased. This is advantageous as a MN count is generated directly after processing the imaging flow cytometry file. This streamlines the process completely whilst maintaining accuracy. Also, by using three different laboratory datasets in the production of the ground truth, application was shown to be accurate for cross-laboratory use, a novelty in this research setting. This allows for the existing ground truth to be used for future MN scoring, allowing for the MN assay to be fully automated.
Genetic Toxicology, DNA damage, Micronucleus, Deep learning, Automation
Faculty of Medicine, Health and Life Sciences