Abstract #2887

Sreenivas M, Gibbs P, Turnbull L, Hunter A

The application of neural networks in MRI has previously been limited due to the requirement for a large number of cases to ensure model robustness. Development of appropriate quantitative parameters, reflecting a radiological interpretation, has also proved problematic. This work utilises an ensemble technique based on bootstrap resampling to improve the final accuracy. Textural, shape, and pharmacokinetic derived parameters are used as inputs into the network.