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 Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Brain tumor imaging has been part of the clinical routine for many years toperform non-invasive detection and grading of tumors. Tumor segmentation is acrucial step for managing primary brain tumors because it allows a volumetricanalysis to have a longitudinal follow-up of tumor growth or shrinkage tomonitor disease progression and therapy response. In addition, it facilitatesfurther quantitative analysis such as radiomics. Deep learning models, inparticular CNNs, have been a methodology of choice in many applications ofmedical image analysis including brain tumor segmentation. In this study, weinvestigated the main design aspects of CNN models for the specific task ofMRI-based brain tumor segmentation. Two commonly used CNN architectures (i.e.DeepMedic and U-Net) were used to evaluate the impact of the essentialparameters such as learning rate, batch size, loss function, and optimizer. Theperformance of CNN models using different configurations was assessed with theBraTS 2018 dataset to determine the most performant model. Then, thegeneralization ability of the model was assessed using our in-house dataset.For all experiments, U-Net achieved a higher DSC compared to the DeepMedic.However, the difference was only statistically significant for whole tumorsegmentation using FLAIR sequence data and tumor core segmentation using T1wsequence data. Adam and SGD both with the initial learning rate set to 0.001provided the highest segmentation DSC when training the CNN model using U-Netand DeepMedic architectures, respectively. No significant difference wasobserved when using different normalization approaches. In terms of lossfunctions, a weighted combination of soft Dice and cross-entropy loss with theweighting term set to 0.5 resulted in an improved segmentation performance andtraining stability for both DeepMedic and U-Net models.

Investigating certain choices of CNN configurations for brain lesion segmentation