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

Fully convolutional deep neural networks have been asserted to be fast andprecise frameworks with great potential in image segmentation. One of the majorchallenges in training such networks raises when data is unbalanced, which iscommon in many medical imaging applications such as lesion segmentation wherelesion class voxels are often much lower in numbers than non-lesion voxels. Atrained network with unbalanced data may make predictions with high precisionand low recall, being severely biased towards the non-lesion class which isparticularly undesired in most medical applications where FNs are moreimportant than FPs. Various methods have been proposed to address this problem,more recently similarity loss functions and focal loss. In this work we trainedfully convolutional deep neural networks using an asymmetric similarity lossfunction to mitigate the issue of data imbalance and achieve much bettertradeoff between precision and recall. To this end, we developed a 3DFC-DenseNet with large overlapping image patches as input and an asymmetricsimilarity loss layer based on Tversky index (using Fbeta scores). We usedlarge overlapping image patches as inputs for intrinsic and extrinsic dataaugmentation, a patch selection algorithm, and a patch prediction fusionstrategy using B-spline weighted soft voting to account for the uncertainty ofprediction in patch borders. We applied this method to MS lesion segmentationbased on two different datasets of MSSEG and ISBI longitudinal MS lesionsegmentation challenge, where we achieved top performance in both challenges.Our network trained with focal loss ranked first according to the ISBIchallenge overall score and resulted in the lowest reported lesion falsepositive rate among all submitted methods. Our network trained with theasymmetric similarity loss led to the lowest surface distance and the bestlesion true positive rate.

ieee access

Asymmetric Loss Functions and Deep Densely Connected Networks for Highly  Imbalanced Medical Image Segmentation: Application to Multiple Sclerosis  Lesion Detection