Landmark localisation in radiographs using weighted heatmap displacement voting

Conference item

We propose a new method for fully automatic landmark localisation using Convolutional Neural Networks (CNNs). Training a CNN to estimate a Gaussian response (“heatmap”) around each target point is known to be effective for this task. We show that better results can be obtained by training a CNN to predict the offset to the target point at every location, then using these predictions to vote for the point position. We show the advantages of the approach, including those of using a novel loss f...

Expand abstract

Authors: Davison, A; Lindner, C; Perry, D; Luo, W; Medical Student Annotation Collaborative

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: Springer, Cham Publisher's website
• Host title: MSKI 2018: Computational Methods and Clinical Applications in Musculoskeletal Imaging
• Series: Lecture Notes in Computer Science
• Journal: MSKI 2018: Computational Methods and Clinical Applications in Musculoskeletal Imaging, Journal website
• Volume: 11404
• Pages: 73-85
• Publication date: 2019-01-09
• Acceptance date: 2018-07-26
• DOI: 10.1007/978-3-030-11166-3_7
• ISSN: 0302-9743
• ISBN: 9783030111663
• Keywords: convolutional neural network (CNN); oting; X-rays; deep learning; paediatrics; Perthes disease; fully convolutional network (FCN)