Abstract
Original language | German |
---|---|
Pages (from-to) | 207-219 |
Journal | Medical image analysis |
Publication status | Published - 2019 |
Cite this
Integrating spatial configuration into heatmap regression based CNNs for landmark localization. / Bischof, Horst; Stern, Darko; Payer, Christian; Urschler, Martin.
In: Medical image analysis, 2019, p. 207-219.Research output: Contribution to journal › Article › Research › peer-review
}
TY - JOUR
T1 - Integrating spatial configuration into heatmap regression based CNNs for landmark localization
AU - Bischof, Horst
AU - Stern, Darko
AU - Payer, Christian
AU - Urschler, Martin
PY - 2019
Y1 - 2019
N2 - In many medical image analysis applications, only a limited amount of training data is available due to the costs of image acquisition and the large manual annotation effort required from experts. Training recent state-of-the-art machine learning methods like convolutional neural networks (CNNs) from small datasets is a challenging task. In this work on anatomical landmark localization, we propose a CNN architecture that learns to split the localization task into two simpler sub-problems, reducing the overall need for large training datasets. Our fully convolutional SpatialConfiguration-Net (SCN) learns this simplification due to multiplying the heatmap predictions of its two components and by training the network in an end-to-end manner. Thus, the SCN dedicates one component to locally accurate but ambiguous candidate predictions, while the other component improves robustness to ambiguities by incorporating the …
AB - In many medical image analysis applications, only a limited amount of training data is available due to the costs of image acquisition and the large manual annotation effort required from experts. Training recent state-of-the-art machine learning methods like convolutional neural networks (CNNs) from small datasets is a challenging task. In this work on anatomical landmark localization, we propose a CNN architecture that learns to split the localization task into two simpler sub-problems, reducing the overall need for large training datasets. Our fully convolutional SpatialConfiguration-Net (SCN) learns this simplification due to multiplying the heatmap predictions of its two components and by training the network in an end-to-end manner. Thus, the SCN dedicates one component to locally accurate but ambiguous candidate predictions, while the other component improves robustness to ambiguities by incorporating the …
M3 - Artikel
SP - 207
EP - 219
JO - Medical image analysis
JF - Medical image analysis
SN - 1361-8415
ER -