Abstract
Hip fractures are a major cause of death and disability for older people and are one of the costliest treatments for the NHS (£1billion annually). Given the high mortality rates ( 9% within 30 days; 30% within 1 year), any improvement in classification, and hence treatment, will have significant benefits. Hip fractures are classified using a standard classification system. However, there is wide varia- tion between hospitals regarding who performs the classification. As classification influences the chosen treatment, this can affect patient outcomes. We are develop- ing a machine learning based method to automatically classify hip fractures using X-rays, with the eventual aim of standardising classification across the NHS.
There are two stages in our proposed classification process: automatically recognise the hip joints in the X-ray image; and then classify the hip fracture given this selected region. Using a GPU and caffe, we have trained a fully convolutional network to automatically locate the hip joint, scoring an intersection-over-union above 0.8 (>0.5 is correct) for 93% of the test set. We are currently training a convolutional neural network to automatically classify the type of fracture, us- ing a dataset of X-ray images that have been hand-labelled by a panel of experts. Details of our methodology and results of our study will be given.
Original language | English |
---|---|
Publication status | Published - 2018 |
Event | HPC Symposium 2018 - University of Bath, Bath, UK United Kingdom Duration: 6 Jun 2018 → 6 Jun 2018 |
Conference
Conference | HPC Symposium 2018 |
---|---|
Country/Territory | UK United Kingdom |
City | Bath |
Period | 6/06/18 → 6/06/18 |