Imperial College London
Portrait Picture

Professor Anand Devaraj

Faculty of Medicine, National Heart & Lung Institute

Professor of Practice (Thoracic Radiology)
 
 
 
//

Contact

 

anand.devaraj Website

 
 
//

Location

 

South BlockRoyal Brompton Campus

//

Summary

 

Publications

Citation

BibTex format

@unpublished{Folgoc:2020,
author = {Folgoc, LL and Baltatzis, V and Alansary, A and Desai, S and Devaraj, A and Ellis, S and Manzanera, OEM and Kanavati, F and Nair, A and Schnabel, J and Glocker, B},
publisher = {arXiv},
title = {Bayesian sampling bias correction: training with the right loss function},
url = {http://arxiv.org/abs/2006.13798v1},
year = {2020}
}
Download

RIS format (EndNote, RefMan)

TY  - UNPB
AB  - We derive a family of loss functions to train models in the presence ofsampling bias. Examples are when the prevalence of a pathology differs from itssampling rate in the training dataset, or when a machine learning practionerrebalances their training dataset. Sampling bias causes large discrepanciesbetween model performance in the lab and in more realistic settings. It isomnipresent in medical imaging applications, yet is often overlooked attraining time or addressed on an ad-hoc basis. Our approach is based onBayesian risk minimization. For arbitrary likelihood models we derive theassociated bias corrected loss for training, exhibiting a direct connection toinformation gain. The approach integrates seamlessly in the current paradigm of(deep) learning using stochastic backpropagation and naturally with Bayesianmodels. We illustrate the methodology on case studies of lung nodule malignancygrading.
AU  - Folgoc,LL
AU  - Baltatzis,V
AU  - Alansary,A
AU  - Desai,S
AU  - Devaraj,A
AU  - Ellis,S
AU  - Manzanera,OEM
AU  - Kanavati,F
AU  - Nair,A
AU  - Schnabel,J
AU  - Glocker,B
PB  - arXiv
PY  - 2020///
TI  - Bayesian sampling bias correction: training with the right loss function
UR  - http://arxiv.org/abs/2006.13798v1
UR  - http://hdl.handle.net/10044/1/80358
ER  -
Download