Predictive modelling and deep learning for quantifying human health

Machine learning and deep learning techniques have emerged as powerful tools for addressing complex challenges across diverse domains. These methodologies are powerful because they extract patterns and insights from large and complex datasets, automate decision-making processes, and continuously improve over time. They enable us to observe and quantify patterns in data that a normal human would not be able to capture, leading to deeper insights and more accurate predictions. This dissertation presents two research papers that leverage these methodologies to tackle distinct yet interconnected problems in neuroimaging and computer vision for the quantification of human health. The first investigation, “Age prediction using resting-state functional MRI,” addresses the challenge of understanding brain aging. By employing the Least Absolute Shrinkage and Selection Operator (LASSO) on resting-state functional MRI (rsfMRI) data, we identify the most predictive correlations related to brain age. Our study, involving a cohort of 176 healthy volunteers, establishes a reference model that significantly reduces prediction errors and identifies abnormal aging patterns, particularly within the Default Mode Network (DMN). This study identifies 39 predictive correlations and achieves a leave-one-out mean absolute error of 2.48 years. Remarkably, our normal reference model attains the lowest prediction error among published models evaluated on adult subjects of almost all ages, highlighting correlations predictive of normal aging. The implications of this work extend to early detection of neurodegenerative diseases, providing a non-invasive method to identify abnormal brain aging before cognitive symptoms manifest, potentially allowing for earlier interventions and personalized treatment strategies. The second investigation, “Skin feature point tracking using deep feature encodings,” explores advancements in computer vision for health monitoring applications. We propose a novel pipeline using a convolutional stacked autoencoder to track facial and skin features, which are crucial for accurate heart rate estimation in ballistocardiography and motor degradation analysis in Parkinson’s disease. Our method achieves tracking errors as low as 0.6-3.3 pixels, outperforming traditional algorithms like SIFT, SURF, and LK in almost all scenarios. Additionally, our approach is the only one that did not diverge and demonstrated superior performance compared to the latest state-of-the-art transformer for feature matching– Omnimotion, especially under conditions of large motion. The implications of this work extend to improving non-invasive health monitoring systems, offering more accurate tools for tracking subtle motor and cardiovascular changes, and enabling early diagnosis and precise monitoring of diseases like Parkinson’s and heart conditions. Together, these studies demonstrate the impact of predictive modeling and deep learning on advancing our understanding and capabilities in neuroimaging and computer vision. This is achieved by advancing neuroimaging by enabling the early detection of abnormal brain aging via functional MRI, while also improving computer vision capabilities for tracking subtle changes in motor functions, offering more accurate, non-invasive tools for diagnosing, quantifying, and monitoring neurodegenerative and cardiovascular diseases.