Full Paper (6 pages)
Official IEEE published version:
DOI: 10.1109/CBMI.2019.8877402
Author's version: PDF
(1.7MB)
Deep neural networks are widely used in the segmentation and classification of medical images. However, little work has addressed the prediction of shapes based on populationdata over time as a regression problem. In this paper, we introduce a regressive convolutional neural network for landmark-based shape prediction. Unlike the conventional CNN model, the proposed network takes the input of a target age, and outputs the corresponding shape for that age. Experimental results demonstrate the effectiveness of the proposed ShapeNet to predict corpus callosum and mandible shapes with correct topology and accurate fitting that matches real-world scenarios. The proposed ShapeNet can predict the shape variation of high dimensional and nonlinear data, which is often critical to understanding the processes that change the shape of anatomy in biology and medical fields.
In Proceedings of the Seventeenth International Conference on Content-Based Multimedia Indexing (CBMI 2019), Dublin, Ireland, September 2019.
Back to Brian Davison's publications