Label stability in multiple instance learning

TY - GEN

T1 - Label stability in multiple instance learning

AU - Cheplygina, Veronika

AU - Sørensen, Lauge

AU - Tax, David M. J.

AU - de Bruijne, Marleen

AU - Loog, Marco

N1 - Conference code: 18

PY - 2015

Y1 - 2015

N2 - We address the problem of instance label stability in multiple instance learning (MIL) classifiers. These classifiers are trained only on globally annotated images (bags), but often can provide fine-grained annotations for image pixels or patches (instances). This is interesting for computer aided diagnosis (CAD) and other medical image analysis tasks for which only a coarse labeling is provided. Unfortunately, the instance labels may be unstable. This means that a slight change in training data could potentially lead to abnormalities being detected in different parts of the image, which is undesirable from a CAD point of view. Despite MIL gaining popularity in the CAD literature, this issue has not yet been addressed. We investigate the stability of instance labels provided by several MIL classifiers on 5 different datasets, of which 3 are medical image datasets (breast histopathology, diabetic retinopathy and computed tomography lung images).We propose an unsupervised measure to evaluate instance stability, and demonstrate that a performance-stability trade-off can be made when comparing MIL classifiers.

AB - We address the problem of instance label stability in multiple instance learning (MIL) classifiers. These classifiers are trained only on globally annotated images (bags), but often can provide fine-grained annotations for image pixels or patches (instances). This is interesting for computer aided diagnosis (CAD) and other medical image analysis tasks for which only a coarse labeling is provided. Unfortunately, the instance labels may be unstable. This means that a slight change in training data could potentially lead to abnormalities being detected in different parts of the image, which is undesirable from a CAD point of view. Despite MIL gaining popularity in the CAD literature, this issue has not yet been addressed. We investigate the stability of instance labels provided by several MIL classifiers on 5 different datasets, of which 3 are medical image datasets (breast histopathology, diabetic retinopathy and computed tomography lung images).We propose an unsupervised measure to evaluate instance stability, and demonstrate that a performance-stability trade-off can be made when comparing MIL classifiers.

U2 - 10.1007/978-3-319-24553-9_66

DO - 10.1007/978-3-319-24553-9_66

M3 - Article in proceedings

SN - 978-3-319-24552-2

T3 - Lecture notes in computer science

SP - 539

EP - 546

BT - Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015

A2 - Navab, Nassir

A2 - Hornegger, Joachim

A2 - Wells, William M.

A2 - Frangi, Alejandro F.

PB - Springer

T2 - International Conference on Medical Image Computing and Computer Assisted Intervention 2015

Y2 - 5 October 2015 through 9 October 2015

ER -