DEFN: Dual-Encoder Fourier Group Harmonics Network for Three-Dimensional Indistinct-Boundary Object Segmentation

TL;DR

This paper introduces DEFN, a novel neural network architecture with stochastic data augmentation and dynamic loss weighting, achieving state-of-the-art results in 3D indistinct-boundary object segmentation in medical imaging.

Contribution

The paper presents DEFN, a dual-encoder Fourier group harmonics network with stochastic defect injection and dynamic weight composing loss, advancing 3D medical image segmentation.

Findings

Achieves state-of-the-art performance on OIMHS dataset.

Effective in handling indistinct boundary medical objects.

Enhances feature recognition and noise filtration.

Abstract

The precise spatial and quantitative delineation of indistinct-boundary medical objects is paramount for the accuracy of diagnostic protocols, efficacy of surgical interventions, and reliability of postoperative assessments. Despite their significance, the effective segmentation and instantaneous three-dimensional reconstruction are significantly impeded by the paucity of representative samples in available datasets and noise artifacts. To surmount these challenges, we introduced Stochastic Defect Injection (SDi) to augment the representational diversity of challenging indistinct-boundary objects within training corpora. Consequently, we propose the Dual-Encoder Fourier Group Harmonics Network (DEFN) to tailor noise filtration, amplify detailed feature recognition, and bolster representation across diverse medical imaging scenarios. By incorporating Dynamic Weight Composing (DWC) loss dynamically adjusts model's focus based on training progression, DEFN achieves SOTA performance on the OIMHS public dataset, showcasing effectiveness in indistinct boundary contexts. Source code for DEFN is available at: https://github.com/IMOP-lab/DEFN-pytorch.