Computer Vision News - February 2023

41 Stine Hansen learning methodology that leverages automatically generated supervoxels in various ways to exploit the structural information in the images. Specifically, I explored few-shot learning-based solutions to organ segmentation and unsupervised approaches for lung tumor segmentation. In the work on few-shot learning-based organ segmentation, supervoxels are used to generate pseudo-labels for self-supervised training. Additionally, to obtain a model that is robust to the typically large and inhomogeneous background class, a novel anomaly detection-inspired classifier is proposed to ease the modelling of the background (Figure 1). Further, to encourage the resulting segmentation maps to respect edges defined in the input space, a supervoxel-informed feature refinement module is proposed to refine the embedded feature vectors during inference (Figure 2). The module automatically identifies areas requiring refinement based on the predictive uncertainty. The work on unsupervised tumor segmentation explores the opportunity of performing clusteringonapopulation-level (asopposedtopatient-level) inorder toprovidethealgorithm with as much information as possible. To facilitate this population-level across-patient clustering, supervoxel representations are exploited to reduce the number of samples, and thereby the computational cost. Within the proposed segmentation framework, five variations of classical clustering algorithms are investigated to evaluate their sensitivity to tumor size and voxel noise, in addition to analyzing the type of segmentation mistakes they are prone to making and quantifying their associated benefit of the population-level clustering. Insummary, through theworkofmy thesis I havedemonstrated that supervoxelsareversatile tools for leveraging structural information in medical images when training segmentation models with limited supervision. Figure 2: Few-shot organ segmentation with and without supervoxel-informed feature refinement. Red: Liver, Blue: Left kidney, Cyan: Spleen.