STOpFrac

 
  • UoM Logo
  • CMFT Logo
  • Optasia Medical

STOpFrac is a collaboration between the University of Manchester, Central Manchester NHS Foundation Trust, and Optasia Medical, with input from the National Osteoporosis Society. The aim is to develop a system that can automatically analyse a variety of clinical images, segment the vertebrae, and identify any abnormalities so that they can be brought to clinical attention.

STOpFrac is funded by a £866K Project Development Award from the National Institue for Health Research Invention for Innovation programme, grant no. II-LB_0216-20009. This project continues on from the HICF-funded project An Automated Tool to Identify Vertebral Fractures in Various Imaging Modalities. For details of each project, see:

STOpFrac: Software Tool for Opportunistic diagnosis of vertebral Fractures
HICF: An Automated Tool to Identify Vertebral Fractures in Various Imaging Modalities.

Osteoporosis is a common skeletal disorder defined by a reduction in bone mineral density. It increases lifetime risk of fragility fractures, typically in the wrists, hips and vertebrae. The vertebral fractures often occur earlier in the course of the disease. Identifying them allows preventative treatments that reduce the risk of a subsequent, much more serious, hip fracture. However, only about one third of vertebral fractures present on clinical images come to clinical attention, often because the images were acquired for other purposes. The potential utility of a computer-aided system that can identify vertebral fractures is therefore considerable.

I have applied the Random Forest Regression Voting Constrained Local Model (RFRV-CLM) to produce high-resolution segmentations of vertebral bodies, allowing subsequent shape measurement and classification, as described in my paper at MICCAI CSI 2014. These models require an initialisation in order to correctly fit an image, which is problematic in spinal images if the number of vertebrae present is unknown. Therefore, I have developed a novel initialisation method based on multiple, independent Random Forest regressors, as described in my paper at MICCAI CSI 2015. Finally, I have developed a system that can identify the spinal mid-plane in 3D CT volumes, allowing fully automatic application of the initialisation and segmentation algorithms - this was presented at MICCAI CSI 2016 in Athens.

Automatic vertebra annotation

The project has met with considerable success. My paper at MICCAI CSI 2014 won an honorable mention in the best paper competition, and we won the Best Poster Award at the International Bone Densitometry Workshop 2014 [abstract, poster]. My paper at MICCAI CSI 2016 was runner-up in the best paper competition. Optasia Medical, our industrial collaborators, won the BioNow Healthcare Product of the year award for their work on the project.

The project had also attracted attention in the press, and has featured in articles in MedicalXpress, The Business Desk, News Medical, Science Daily, Health Canal, Medical News Today, MDLinx, The Engineer, The Daily Mail, and the web page of the International Society for Clinical Densitometry (ISCD). It also appeared in the MAHSC newsletter and as a case study on the MAHSC website.

ASPIRE™

 

The machine learning algorithms developed through this project are used in the AVERT software that our industrial collaborators Optasia Medical Ltd use in their ASPIRE™ service. The software is currently in use at the University of Manchester and the University of Sheffield. Potential users of the AVERT software or the ASPIRE™ service should contact Optasia Medical to discuss their requirements.

AVERT Software

Previous Research Projects

 

Pages describing some of my previous research projects are hosted on my personal website on the TINA server. There, you will find descriptions of: