27 Jun 2017

Determination of Knee Osteoarthritis Severity using Deep learning

The motivation behind this project is to provide the rheumatologists with an automatic analysis of the severity of Osteoarthritis. This can be used to assess the effectiveness of therapeutic interventions. For this research, the dataset used was provided to me by MOST and The Osteoarthritis Initiative, UCLA. These datasets together have about 1500 images along with their annotations. Although more images could be obtained from the local hospitals, they will have to be annotated by rheumatologists and that will be a costly task. I decided to use transfer learning and also make use of data augmentation techniques to make up for the lack of data. I am currently working on developing a model inspired by the latest ILSVRC model by NASNET.

The images are of very high quality and hence do not require much of preprocessing except that of reducing the dimensions and conversion from DICOM format to Jpg or Png using the Pydicom package.