NASA
Research type
Research Study
Full title
NASA: Novel Applications for Sarcoma Assessment
IRAS ID
326856
Contact name
Matthew Marzetti
Contact email
Sponsor organisation
Leeds Teaching Hospitals NHS Trust
Duration of Study in the UK
3 years, 5 months, 31 days
Research summary
Research Summary:
Our research aim is to improve the way of deciding whether a lump in soft tissue such as fat or muscle is a type of cancer called a soft tissue sarcoma, or if it is benign using artificial intelligence (AI).
Soft tissue sarcomas are a type of cancer that can appear anywhere in the body where there is soft tissue such as muscle or fat. While sarcomas are rare, benign lumps in soft tissue are common and it is currently very difficult to tell the difference between the two using imaging. This means many patients with benign masses are referred for painful biopsies and waiting lists for biopsies are long due to the large diagnostic workload.
This research aims to develop an AI algorithm that can differentiate between benign and malignant soft tissue masses. While we can develop an algorithm using existing routine data we would like to investigate if adding quantitative MR images could make it more accurate.
We will ask patients who are already having a scan for sarcoma if we can take extra MR images. We will use these images to provide extra information to the AI. The extra images will add a maximum of 10 minutes to the patients’ standard MRI scan, meaning patients will not need to make an extra trip or undergo any extra procedures. Study participants will not need to receive MR contrast as part of this research. The extra images will not be used to make a diagnosis during this research. We will also ask a small subset of patients if they would be willing to come for a second scan so that we can see how reliable our measurements are, but this will be entirely optional.
Lay summary of study results:
Soft tissue tumours (STTs) are lumps that develop in the soft tissues of the body. Most are harmless, but some are cancerous or have the potential to behave aggressively. It can be difficult to tell the difference between these tumours using medical imaging alone, and patients may therefore need an invasive biopsy or surgery to establish a diagnosis.
This research investigated whether information extracted from magnetic resonance imaging (MRI) could be used to help distinguish between benign and potentially aggressive soft tissue tumours. The study used a technique called radiomics, which uses computer algorithms to measure features within medical images that may not be readily visible to the human eye.
The research involved several stages. A review of previous research into the use of artificial intelligence for diagnosing soft tissue tumours was undertaken to identify gaps in the evidence. Patients, radiologists, pathologists and other clinical stakeholders were also involved in the research to help ensure that the approach addressed clinically relevant problems and that the evaluation reflected the needs of patients and healthcare professionals.
MRI data from 951 patients were used to develop the computer models. The models were then tested using a further 158 patients who were studied prospectively, as well as an independent dataset of 162 patients from other institutions. The models used information from MRI scans, including radiomics measurements, as well as clinical and radiologist-assessed information.
The radiomics-based model was able to distinguish benign from potentially aggressive tumours with good performance. When tested on the prospective dataset, the model correctly identified 89% of potentially aggressive tumours, while correctly identifying 55% of benign tumours. Performance was also good in the independent dataset, where 88% of potentially aggressive tumours and 88% of benign tumours were correctly identified. Performance was lower for some types of benign tumour, particularly myxoid tumours.
These findings suggest that analysing MRI scans using radiomics could potentially help identify benign soft tissue tumours and reduce the number of patients requiring further investigation, while maintaining a high level of sensitivity for potentially aggressive tumours. However, the research does not demonstrate that the system is ready to replace current clinical assessment or biopsy. Further research, including testing in larger and more diverse patient populations and evaluation alongside clinical practice, is needed before such an approach could be considered for routine clinical use.
REC name
Yorkshire & The Humber - South Yorkshire Research Ethics Committee
REC reference
23/YH/0151
Date of REC Opinion
4 Jul 2023
REC opinion
Favourable Opinion