Development of an AI-driven algorithm based on acute diagnostic CT brain scans to predict recovery from intracerebral haemorrhage

Abstract

Intracerebral haemorrhage (ICH) is a form of stroke caused by the rupture of a blood vessel in the brain, and is associated with high rates of disability and death. One key determinant of functional recovery after ICH is the integrity of the corticospinal tract (CST), the principal white matter pathway responsible for fine motor control. This thesis investigates how CST integrity, assessed using routine diagnostic imaging, can be used to predict functional outcomes after ICH, particularly in the context of surgical haematoma evacuation. Firstly we present a standardised manual tool for inferring CST integrity from non-contrast CT, and we demonstrate the strong association between CST integrity and motor function in the MISTIE III clinical trial dataset when used by an expert. However, a validation of the tool revealed high interobserver variability. In response, two deep learning models were developed and compared: one trained on manual CST segmentations, and another trained using diffusion tensor imaging tractography masks of the CST. While both models successfully predicted motor outcomes in the MISTIE III dataset, only the tractography-based model generalised to the CLEAR III trial dataset, demonstrating improved robustness. To explore the broader functional consequences of lesion location, lesion-to-symptom mapping revealed distinct areas of the brain associated with motor outcome, that evolved over time, and highlighted differences between medically managed and surgical patients. Analysis of the disconnectome identified CST disconnection as the strongest independent predictor of chronic motor impairment, and revealed differences between medical and surgical patients that suggest surgery may put compensatory pathways at risk. Finally, the role of CST integrity in modulating the effect of surgery was tested. CST integrity was shown not to predict differential benefit from surgery, however haematoma proximity to the CST significantly influenced outcome. Patients with haematomas within 1-3mm of the CST had significantly worse motor outcome after surgery, the CST may be sustaining injury during surgical intervention. These studies demonstrate that CST integrity can be assessed using CT imaging, predicts long-term motor recovery, and may inform surgical risk in ICH. Our findings highlight the potential of personalised imaging based prognostication and show the need for future studies into CST-oedema interactions and early surgical intervention.

Date of Award	30 Jun 2025
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Timothy Cootes (Main Supervisor), Hiren Patel (Co Supervisor) & Adrian Parry-Jones (Co Supervisor)