Machine learning methods for AI-enhanced DCE-MRI breast cancer diagnosis

Abstract

Breast cancer remains one of the most prevalent and challenging diseases affecting women globally. Early and accurate diagnosis plays a pivotal role in improving patient outcomes and reducing mortality rates. This project aims to improve the current state of the art by employing several deep learning methodologies to be used as diagnostic support in Dynamic Contrast Enhanced Magnetic Resonance Imaging (DCE-MRI). The work has been carried out as part of an InnovateUK project named “Intelliscan” (project reference: 104192), funded by UK Research and Innovation. UK Research and Innovation did not have any involvement in the study design, or the collection, analysis, and interpretation of data. The data collection was carried out in collaboration with consultant radiologist Dr Naveed Altaf (North Tees and Hartlepool NHS Foundation Trust) and Dr Susann Wolfram (Teesside University). The research begins by providing an overview of neural networks, including structure, activation, regularization, training, architectures, loss functions, and performance metrics for model evaluation. The first contribution to knowledge is provided as the diagnostic process is then presented from a clinician’s point of view, along with empirical evidence of the subjectivity of the process prevents the application of ground truth data for the development of algorithms in this area. The core contributions of this thesis lie in the development of several deep learning methodologies that are aimed at reducing human errors and increasing the speed of the diagnosis in clinical settings. These methodologies include: the first lesion detection algorithm based on unsupervised deep learning, which matches the performance of the current state of the art, while not relying on manually annotated data, significantly lowering the cost of research in the field; the development of a state of the art deep learning methodology to segment the organs within the chest wall from the breast; a novel application of deep learning in lesion morphology characterisation. Overall, this thesis contributes to advancing the state-of-the-art in breast DCE-MRI diagnosis by introducing innovative deep learning methodologies that offer enhanced accuracy, efficiency, and interpretability.