Automated Detection of Spinal Disorders Using MRI Scans and Radiology Reports

Accurate and timely evaluations of magnetic resonance imaging (MRI) scans of the spine are necessary to detect cancerous spine tumours and monitor progressive degenerative conditions. Back pain affects 80% of people in the UK over their lifetime, yet the causes of back pain are not fully understood. Delayed diagnosis or treatment of spinal conditions has been associated with worse outcomes-pain and disability for the patient and more expensive and complicated treatments at later stages of disease.

Advancements in computer vision and medical imaging have led to the development of tools that can produce automated vertebral segmentations and radiological gradings. However, these systems are not meant to be used as diagnosis tools and presently only generate qualitative evaluations of disease progression. Currently, there is no tool in clinical practice that can replicate the function of a radiologist for any spinal disorder.

This project falls within the EPSRC healthcare technologies research area. We aim to develop a system that is closer to the way radiologists work by incorporating both vision and text data. Radiology reports are rich with details that are missing from tabular annotations traditionally used in medical imaging tasks, which will reduce the need for manual annotation of the MRIs and preserve complicated and overlapping symptomologies often present in clinical practice. We will explore using the reports to create normalised structured labels, as well as training bi-modal vision-language models (VLM) on a newly compiled dataset from the National Consortium of Intelligent Medical Imaging (NCIMI), which includes around 5,000 spinal MRI scans paired with the accompanying radiology reports.

Specifically, the aims of the research are:

1. Structured supervision from reports: We will use new, multi-centre data from NCIMI to detect spinal cancer. The data contain many paired image-report samples from patients with cancer. We will use the reports to create structured labels similar to those generated by manual annotation and use descriptions of primary or secondary incidence of cancer from the reports to identify corresponding sites in the images. We will set aside data from a single centre to validate on unseen data and demonstrate usefulness in real clinical settings.

2. Visual-language modelling: We will adapt vision-language models to use the free-text reports directly, without restructuring them to tabular form. Using the reports as direct inputs to the model should result in better performance in the classification of spinal conditions than using images alone. Details from the reports may also help us pinpoint the specific location of abnormalities in the images without the need of manual annotation. Incorporating text inputs would ideally allow for an interactive system that can answer user questions with natural language responses when prompted.

3. Quantitative measurements: We will develop more quantitative biomarkers for spinal conditions. Grading systems currently used by radiologists and incorporated into SpineNet are qualitative and subject to large inter-rater variability. Getting objective metrics of disease progression from accurately predicted segmentations can define more precise phenotypes of disease stages, which would be useful in supporting clinical studies and drug trials.

The proposed studies in combination will form a system that can take both image and text inputs to detect abnormalities in the spine and generate narrative responses to questions posed by the user, with quantitative biomarkers that can be linked to disease phenotypes. The successful development and deployment of such a tool would enable health practitioners to deliver care at greater efficiency, reduce costs and lead to better patient outcomes.

In the same way that bioinformatics has transformed genomic research and clinical practice, health data science will have a dramatic and lasting impact upon the broader fields of medical research, population health, and healthcare delivery. The beneficiaries of the proposed training programme, and of the research that it delivers and enables, will include academia, industry, healthcare, and the broader UK economy.

Academia: Graduates of the training programme will be well placed to start their post-doctoral careers in leading academic institutions, engaging in high-impact multi-disciplinary research, helping to build training and research capacity, sharing their experience within the wider academic community.

Industry: Partner organisations will benefit from close collaboration with leading researchers, from the joint exploration of research priorities, and from the commercialisation of arising intellectual property. Other organisations will benefit from the availability of highly-qualified graduates with skills in big health data analytics.

Healthcare: Healthcare organisations and patients will benefit from the results of enabled and accelerated health research, leading to new treatments and technologies, and an improved ability to identify and evaluate potential improvements in practice through the analysis of real-world health data.

Economy: The life sciences sector is a key component of the UK economy. The programme will provide partner companies with direct access to leading-edge research. Graduates of the programme will be well-qualified to contribute to economic growth - supporting health research and the development of new products and services - and will be able to inform policy and decision making at organisational, regional, and national levels.