Advanced image analysis of mitochondria in biomedical science problems

Early and accurate disease detection remains a major challenge in modern medicine and is essential for advancing drug development and improving treatment outcomes. Reliable assessment of cellular health is particularly difficult. Mitochondria, central organelles in energy production, are emerging as powerful biomarkers, with dysfunction linked to conditions including neurodegenerative disorders, cancer and reproductive challenges. However, current diagnostic methods are often invasive, time-consuming and lack standardisation, highlighting the need for precise, scalable and non-invasive approaches to evaluate mitochondrial health. This thesis presents a label-free approach for assessing mitochondrial networks using autofluorescence microscopy combined with machine learning. High-resolution imaging was performed by leveraging the endogenous fluorescence of NAD(P)H, a key metabolic coenzyme predominantly localised in mitochondria. To analyse these images, AutoMitoNetwork was developed as a software tool that extracts quantitative morphology, intensity and texture features from mitochondrial networks. These features were integrated into machine learning models to classify cell groups and detect mitochondrial dysfunction. We also developed SMIAL, an open-source tool with a graphical user interface that allows users to build machine learning pipelines without coding, simplifying multidimensional analysis and supporting broader accessibility. In addition, we explored deep learning for cell classification using brightfield microscopy, a widely accessible imaging technique. Despite limitations in contrast and subcellular resolution, the models accurately identified mitochondrial dysfunction. To enhance clinical relevance, we focused on improving model generalisability across varied imaging conditions, ensuring robustness in real-world applications. The methodologies developed in this thesis are being applied across three ongoing clinical contexts: oocyte quality assessment, mitochondrial disease diagnosis and bioenergetics research in motor neurone disease. These applications highlight the translational potential of the tools introduced. By advancing label-free imaging and analysis of mitochondrial networks, this work contributes to more accessible and standardised diagnostics and deepens our understanding of mitochondrial morphology as a biomarker, opening new avenues for early disease detection and reinforcing the central role of mitochondria in health and disease.