English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Deep convolutional neural networks (DCNN) are currently ubiquitous in medicalimaging. While their versatility and high quality results for common imageanalysis tasks including segmentation, localisation and prediction isastonishing, the large representational power comes at the cost of highlydemanding computational effort. This limits their practical applications forimage guided interventions and diagnostic (point-of-care) support using mobiledevices without graphics processing units (GPU). We propose a new scheme thatapproximates both trainable weights and neural activations in deep networks byternary values and tackles the open question of backpropagation when dealingwith non-differentiable functions. Our solution enables the removal of theexpensive floating-point matrix multiplications throughout any convolutionalneural network and replaces them by energy and time preserving binary operatorsand population counts. Our approach, which is demonstrated using afully-convolutional network (FCN) for CT pancreas segmentation leads to morethan 10-fold reduced memory requirements and we provide a concept forsub-second inference without GPUs. Our ternary approximation obtains highaccuracies (without any post-processing) with a Dice overlap of 71.0% that arestatistically equivalent to using networks with high-precision weights andactivations. We further demonstrate the significant improvements reached incomparison to binary quantisation and without our proposed ternary hyperbolictangent continuation. We present a key enabling technique for highly efficientDCNN inference without GPUs that will help to bring the advances of deeplearning to practical clinical applications. It has also great promise forimproving accuracies in large-scale medical data retrieval.

TernaryNet: Faster Deep Model Inference without GPUs for Medical 3D Segmentation using Sparse and Binary Convolutions