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

Medical ultrasound technology is widely used in routine clinical applicationssuch as disease diagnosis and treatment as well as other applications likereal-time monitoring of human tongue shapes and motions as visual feedback insecond language training. Due to the low-contrast characteristic and noisynature of ultrasound images, it might require expertise for non-expert users torecognize tongue gestures. Manual tongue segmentation is a cumbersome,subjective, and error-prone task. Furthermore, it is not a feasible solutionfor real-time applications. In the last few years, deep learning methods havebeen used for delineating and tracking tongue dorsum. Deep convolutional neuralnetworks (DCNNs), which have shown to be successful in medical image analysistasks, are typically weak for the same task on different domains. In manycases, DCNNs trained on data acquired with one ultrasound device, do notperform well on data of varying ultrasound device or acquisition protocol.Domain adaptation is an alternative solution for this difficulty bytransferring the weights from the model trained on a large annotated legacydataset to a new model for adapting on another different dataset usingfine-tuning. In this study, after conducting extensive experiments, weaddressed the problem of domain adaptation on small ultrasound datasets fortongue contour extraction. We trained a U-net network comprises of anencoder-decoder path from scratch, and then with several surrogate scenarios,some parts of the trained network were fine-tuned on another dataset as thedomain-adapted networks. We repeat scenarios from target to source domains tofind a balance point for knowledge transfer from source to target and viceversa. The performance of new fine-tuned networks was evaluated on the sametask with images from different domains.

Transfer Learning for Ultrasound Tongue Contour Extraction with Different Domains