1. Background
We developed an AI tool that can accurately sort conventional MRI brain scans into normal and abnormal, resulting in faster management of patients. In parallel, ultra-low-field MRI scanners have recently emerged, which are far cheaper than conventional high-field MRI, are mobile, and run using a standard electrical socket. King’s has access to several portable MRI systems, which we are using to image patients referred for routine clinical high-field MRI. However, it isn’t clear whether portable MRI scans can be used for automated patient triage.
2. Novelty & Importance
Portable ultra-low-field MRI has the potential to considerably widen access to clinical neuroimaging. In high-income settings, portable MRI scanners could be used for triage of patients in community diagnostic centres or even GP services. In low-and-middle-income settings, portable MRI could facilitate diagnosis of patients with limited access to standard high-field MRI. Integrating deep-learning methods for automated abnormality detection into portable MRI workflows would further accelerate diagnostic decisions.
3. Aims & Objectives
The student will have access to a unique dataset including patients with a range of diagnoses as well as healthy participants, scanned at both ultra-low- and high-field (500+ participants), to explore the practical relevance of clinical ultra-low-field MRI.
Specific objectives include:
• Fine-tune and analytically validate an AI tool for triaging clinical portable MRI brain scans
• Assess whether existing methods for analysis of conventional high-field neuroimaging are suitable for portable ultra-low-field MRI
• Explore image enhancement methods to improve the performance of ultra-low-field MRI for abnormality detection and quantitative analysis
Váša F, Bennallick C, Bourke NJ, Padormo F, Baljer L, Briski U, Cawley P, Arichi T, Wood TC, Lythgoe DJ, Dell’Acqua F, Booth TC, Venkataraman AV, Ljungberg E, Deoni SCL, Moran RJ, Leech R*, Williams SCR* (2025). Ultra-low-field brain MRI morphometry: test-retest reliability and correspondence to high-field MRI. Imaging Neuroscience doi.org/10.1162/IMAG.a.930
Wood DA, Kafiabadi S, Busaidi AA, […], Booth TC. (2022). Deep learning models for triaging hospital head MRI examinations. Medical Image Analysis 78:102391.
Arnold TC, Freeman CW , Litt B, and Stein JM. (2023). Low-field MRI: Clinical promise and challenges. Journal of Magnetic Resonance Imaging 57 (1): 25–44.