Automatic anomaly detection for brain imaging triage and classification (360G-Wellcome-103709_Z_14_Z)

Modern brain imaging generates many thousands of data observations per patient, yet its clinical impact remains determined only by the summary points in a radiologist's verbal report. This information gap shows that we are not fully exploiting all potentially relevant clinical information from this expensively acquired data. Our technology will seek dramatically to reduce this gap within current clinical pathways by applying novel machine-learning algorithms to routine brain imaging data, with the aim of automatically extracting rich, high-dimensional information of clinical utility. The system will deliver automatic quantification of the anomaly of each data point of an image to assist radiological reporting and automated outcome prediction based on disease patterns and an anomaly score for the whole image to aid radiological triage and resource/performance management. Our goal is to demonstrate the feasibility, robustness, clinical, and managerial value of the approach using a large dataset of routine clinical brain imaging, delivering a pilot system translatable into a full clinical product. Merely by adding an inexpensive layer of computation to existing pathways, the system should improve report fidelity and optimise radiological triage and management, while creating a scalable new platform for facilitating big data' approaches to major neurological disorders.