Neural Annotation Refinement: Development of a New 3D Dataset for Adrenal Gland Analysis

The human annotations are imperfect, especially when produced by junior practitioners. Multi-expert consensus is usually regarded as golden standard, while this annotation protocol too expensive to implement in many real-world projects. In study, we propose a method refine annotation, named Neural Annotation Refinement (NeAR). It based on learnable implicit function, which decodes latent vector into represented shape. By integrating the appearance an input of functions, appearance-aware NeAR fixes artefacts. Our demonstrated application adrenal gland analysis. We first show that can repair distorted standards public segmentation dataset. Besides, develop new Adrenal gLand ANalysis (ALAN) dataset with proposed NeAR, where each case consists 3D shape and its diagnosis label (normal vs. abnormal) assigned experts. models trained shapes repaired diagnose glands better than original ones. ALAN will be open-source, 1,594 for diagnosis, serves benchmark medical Code available at https://github.com/M3DV/NeAR .