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

Deep learning models have shown their potential for several applications.However, most of the models are opaque and difficult to trust due to theircomplex reasoning - commonly known as the black-box problem. Some fields, suchas medicine, require a high degree of transparency to accept and adopt suchtechnologies. Consequently, creating explainable/interpretable models orapplying post-hoc methods on classifiers to build trust in deep learning modelsare required. Moreover, deep learning methods can be used for segmentationtasks, which typically require hard-to-obtain, time-consumingmanually-annotated segmentation labels for training. This paper introducesthree inherently-explainable classifiers to tackle both of these problems asone. The localisation heatmaps provided by the networks -- representing themodels' focus areas and being used in classification decision-making -- can bedirectly interpreted, without requiring any post-hoc methods to deriveinformation for model explanation. The models are trained by using the inputimage and only the classification labels as ground-truth in a supervisedfashion - without using any information about the location of the region ofinterest (i.e. the segmentation labels), making the segmentation training ofthe models weakly-supervised through classification labels. The finalsegmentation is obtained by thresholding these heatmaps. The models wereemployed for the task of multi-class brain tumour classification using twodifferent datasets, resulting in the best F1-score of 0.93 for the supervisedclassification task while securing a median Dice score of 0.67$\pm$0.08 for theweakly-supervised segmentation task. Furthermore, the obtained accuracy on asubset of tumour-only images outperformed the state-of-the-art glioma tumourgrading binary classifiers with the best model achieving 98.7\% accuracy.

Weakly-supervised segmentation using inherently-explainable classification models and their application to brain tumour classification