Publication Search Results

Now showing 1 - 2 of 2
  • (2021) Collins, Scott
    Chronic liver diseases including cirrhosis and primary liver cancer are a significant health burden worldwide. Liver cirrhosis is end stage liver injury resulting in a progressive fibrosis phenotype, in which the hepatic architecture is distorted. The most common cause of cirrhosis is chronic liver injury caused by hepatitis B, alcohol related liver disease, hepatitis C or non-alcoholic steatohepatitis. Primary liver cancer is a leading cause of cancer mortality globally and is commonly observed as a progression of liver cirrhosis. Liver injury usually occurs because of immune-mediated or direct injury to the hepatocytes and involves multiple cellular subsets; including hepatic stellate cells, liver adipocytes, liver resident macrophages kupffer cells, endothelial cells and infiltrating immune cells. Injury to these cells result in the release of reactive oxygen species, proinflammatory signals, proliferation-associated cytokines, and the activation of repair pathways. A chronic activation of these signals can result in dysregulation of the normal repair response and generation of a pathogenic fibrotic response. A broadly canonical response, chronic inflammation drives fibrosis and cirrhosis irrespective of liver injury aetiology. The burden of liver disease provides the impetus to pursue the use of representative in vitro models of liver function and responses to injury. Improved 2D and 3D in vitro disease models would enhance our understanding of the causes of liver injury and the development of cirrhosis and primary liver cancer while increasing the efficacy of preclinical drug discovery. Current 2D in vitro assays based on cell lines such as HepG2 that have reduced metabolic capacities compared to primary hepatocytes ex vivo, and the use of primary human hepatocytes suffers from high donor-to-donor variation and only retain in vivo characteristics for a short time ex vivo. The shortcomings of 2D cell culture models have driven the development of 3D cell culture techniques. The advantages of 3D models include replicating the complex attributes of the liver beyond liver specific metabolism, such as increased cell density, organisation, and cell-cell signalling, O2 zonation, as well as the anatomy of the liver lobule and the circulatory system. After a comprehensive review of all the current in vitro models of the liver we hypothesised that a liver organoid cell culture model co-cultured with myofibroblast like hepatic stellate cells can model liver injury. An organoid cell culture is defined as a collection of cells culturing several cell types that develop from stem cells or organ progenitors and self-organise through cell sorting and spatially restricted lineage commitment, similar to organogenesis in vivo. Liver organoids have demonstrated many advantages over conventional in vitro models such as long-term genetic stability, 2D in vivo-like organisation, and maintaining the necessary cellular cross talk and behavioural characteristics of their primary corresponding cells. The focus of this thesis is the application of 3D liver organoids to model and analyse the molecular and cellular effects of liver injury. We established a 3D liver organoid cell culture model from primary mouse tissue and characterised the capacity of these organoids to model liver characteristics in vitro and used this model to define the interactions between organoid hepatocytes and hepatic stellate cells in a co-culture trans-well system. The impact of inflammatory cytokines tumour necrosis factor-α and transformation growth factor-β on this model, as well as other variables such as hypoxia and the anti-fibrosis drug Halofuginone were assessed. Hepatic stellate cell dependent decreases in organoid viability and organoid dependent increases in hepatic stellate cell viability were observed, as well as Halofuginone dependent decreases in hepatic stellate cell viability were also observed. Markers characteristic of liver injury and fibrosis, such as Actn1 and Lamb3 were upregulated in hepatic stellate cells, although collagen expression was downregulated in these cells. Transcriptional profiling revealed a tumour necrosis factor-α mediated apoptotic response in organoids and an inflammatory response in both the organoids and hepatic stellate cells. We concluded that while liver organoids and hepatic stellate cells responded to experimental variables, there were limitations when it came to the cross talk between the cultures in the trans-well system. While apoptotic bodies from the organoids may have stimulated proliferation of hepatic stellate cells, many key genes responsible for liver injury were either not upregulated or were downregulated in co-culture. Electron microscopy analysis of liver organoids showed important ultrastructural changes compared to a whole liver section. Our findings of secreted exosomes, microvilli within the lumen of the organoids, and many ultrastructural features found within liver cells in vivo confirm that our 3D liver organoids closely resemble the liver. We also demonstrated how the use of high-resolution field emission scanning electron microscopy with automated scan resolution can generate a high-resolution ultrastructure map of the whole organoid. This method can also be combined with correlative light electron microscopy for immunofluorescent labelling of proteins of interest using quantum dot nanoparticles. Overall, our 3D organoid model of liver injury had encouraging results and furthering our understanding of pathogenesis of liver fibrogenesis in vitro and the study of novel anti-fibrotic therapeutic agents.

  • (2023) Akter, Nahida
    Glaucoma is a multi-factorial, progressive blinding optic-neuropathy. A variety of factors, including genetics, vasculature, anatomy, and immune factors, are involved. Worldwide more than 80 million people are affected by glaucoma, and around 300,000 in Australia, where 50% remain undiagnosed. Untreated glaucoma can lead to blindness. Early detection by Artificial intelligence (AI) is crucial to accelerate the diagnosis process and can prevent further vision loss. Many proposed AI systems have shown promising performance for automated glaucoma detection using two-dimensional (2D) data. However, only a few studies had optimistic outcomes for glaucoma detection and staging. Moreover, the automated AI system still faces challenges in diagnosing at the clinicians’ level due to the lack of interpretability of the ML algorithms and integration of multiple clinical data. AI technology would be welcomed by doctors and patients if the "black box" notion is overcome by developing an explainable, transparent AI system with similar pathological markers used by clinicians as the sign of early detection and progression of glaucomatous damage. Therefore, the thesis aimed to develop a comprehensive AI model to detect and stage glaucoma by incorporating a variety of clinical data and utilising advanced data analysis and machine learning (ML) techniques. The research first focuses on optimising glaucoma diagnostic features by combining structural, functional, demographic, risk factor, and optical coherence tomography (OCT) features. The significant features were evaluated using statistical analysis and trained in ML algorithms to observe the detection performance. Three crucial structural ONH OCT features: cross-sectional 2D radial B-scan, 3D vascular angiography and temporal-superior-nasal-inferior-temporal (TSNIT) B-scan, were analysed and trained in explainable deep learning (DL) models for automated glaucoma prediction. The explanation behind the decision making of DL models were successfully demonstrated using the feature visualisation. The structural features or distinguished affected regions of TSNIT OCT scans were precisely localised for glaucoma patients. This is consistent with the concept of explainable DL, which refers to the idea of making the decision-making processes of DL models transparent and interpretable to humans. However, artifacts and speckle noise often result in misinterpretation of the TSNIT OCT scans. This research also developed an automated DL model to remove the artifacts and noise from the OCT scans, facilitating error-free retinal layers segmentation, accurate tissue thickness estimation and image interpretation. Moreover, to monitor and grade glaucoma severity, the visual field (VF) test is commonly followed by clinicians for treatment and management. Therefore, this research uses the functional features extracted from VF images to train ML algorithms for staging glaucoma from early to advanced/severe stages. Finally, the selected significant features were used to design and develop a comprehensive AI model to detect and grade glaucoma stages based on the data quantity and availability. In the first stage, a DL model was trained with TSNIT OCT scans, and its output was combined with significant structural and functional features and trained in ML models. The best-performed ML model achieved an area under the curve (AUC): 0.98, an accuracy of 97.2%, a sensitivity of 97.9%, and a specificity of 96.4% for detecting glaucoma. The model achieved an overall accuracy of 90.7% and an F1 score of 84.0% for classifying normal, early, moderate, and advanced-stage glaucoma. In conclusion, this thesis developed and proposed a comprehensive, evidence-based AI model that will solve the screening problem for large populations and relieve experts from manually analysing a slew of patient data and associated misinterpretation problems. Moreover, this thesis demonstrated three structural OCT features that could be added as excellent diagnostic markers for precise glaucoma diagnosis.