Séminaire du 30 octobre 2023

Medical image segmentation with deep learning and evidence reasoning

Ling Huang, Saw Swee Hock School of Public Health, National University of Singapore

Résumé :

The comprehensive integration of deep learning models within clinical practice remains suboptimal, notwithstanding the proliferation of high-performing solutions reported in the literature. A predominant factor hindering widespread adoption pertains to an insufficiency of evidence affirming the reliability of existing models, especially due to the limitations for reasoning with and combining imperfect (imprecise, uncertain, and partial) information.

In this talk, I will introduce my research work on medical image segmentation studies with belief function theory [1][2] and deep learning, specifically focusing on information modeling and fusion based on uncertain evidence. First, a comprehensive overview of prevailing methods proposed to quantify uncertainty inherent in machine learning models developed for various medical image tasks will be provided. Second, I will introduce two evidential classifiers, evidential neural network [3] and radial basis function network [4], and show the effectiveness of belief function theory in uncertainty quantification. Furthermore, the two evidential classifiers were integrated with deep neural networks to construct deep evidential models for lymphoma segmentation. Third, I will introduce a multimodal medical image fusion framework that taking into account the reliability of each MR image source when performing different segmentation tasks using mass functions and contextual discounting, and show how the proposed framework can be used to improve the reliability and explainablity of the results. At the end, some possible research work will be given.