Figure_3.tif (1.71 MB)
Download file

Random survival forest analysis of biomarker image feature distributions.

Download (0 kB)
posted on 2014-09-22, 02:55 authored by Claudia Bühnemann, Simon Li, Haiyue Yu, Harriet Branford White, Karl L. Schäfer, Antonio Llombart-Bosch, Isidro Machado, Piero Picci, Pancras C. W. Hogendoorn, Nicholas A. Athanasou, J. Alison Noble, A. Bassim Hassan

An overview of the imaging, the RSF survival analysis algorithm and validation approach. Single cell features are combined into patient features by estimating the probability distribution (PDF) for each feature, and taking measurements of each distribution at 100 points. Each RSF is used to analyse all patients, with prognostic features identified. The use of bagging in each RSF means error rate estimates should be unbiased, and this is verified using randomised cross-validation. This procedure also allows the variability in performance of the algorithm to be simulated without requiring an additional dataset.