Probabilistic Spatial Regression using a Deep Fully Convolutional Neural Network

Slabaugh, G.G., Knapp, K. & Al-Arif, S. M. (2017). Probabilistic Spatial Regression using a Deep Fully Convolutional Neural Network. Paper presented at the British Machine Vision Conference 2017, 4-7 Sep 2017, London, UK.

[img]
Preview
Text - Published Version
Download (3MB) | Preview

Abstract

Probabilistic predictions are often preferred in computer vision problems because they can provide a confidence of the predicted value. The recent dominant model for computer vision problems, the convolutional neural network, produces probabilistic output for classification and segmentation problems. But probabilistic regression using neural networks is not well defined. In this work, we present a novel fully convolutional neural network capable of producing a spatial probabilistic distribution for localizing image landmarks. We have introduced a new network layer and a novel loss function for the network to produce a two-dimensional probability map. The proposed network has been used in a novel framework to localize vertebral corners for lateral cervical Xray images. The framework has been evaluated on a dataset of 172 images consisting 797 vertebrae and 3,188 vertebral corners. The proposed framework has demonstrated promising performance in localizing vertebral corners, with a relative improvement of 38% over the previous state-of-the-art.

Item Type: Conference or Workshop Item (Paper)
Additional Information: Copyright the authors, 2017.
Subjects: Q Science > QA Mathematics
Q Science > QA Mathematics > QA75 Electronic computers. Computer science
R Medicine > RC Internal medicine > RC0321 Neuroscience. Biological psychiatry. Neuropsychiatry
Divisions: School of Informatics > Department of Computing
URI: http://openaccess.city.ac.uk/id/eprint/17827

Actions (login required)

View Item View Item

Downloads

Downloads per month over past year

View more statistics