A Novel Convolutional-Free Method for 3D Medical Imaging Segmentation
Canxuan Gang
TL;DR
This work tackles the challenge of 3D medical image segmentation by eliminating convolutional inductive bias in favor of a fully transformer-based architecture to better model global context. It introduces a patch-based, convolution-free encoder–decoder framework and a novel thin–thick adaptation loss that enables accurate thin-slice segmentation using thick-slice annotations, complemented by a new thin-slice multi-semantic benchmark for NCCT brain hemorrhage. The approach reportedly outperforms traditional CNNs and hybrid architectures, with emphasized domain adaptation between thick and thin slices and robust multi-label segmentation. The contributions—model design, joint loss, and a public thin-slice dataset—advance convolution-free segmentation in medical imaging and hold promise for improved diagnostic and surgical planning outcomes.
Abstract
Segmentation of 3D medical images is a critical task for accurate diagnosis and treatment planning. Convolutional neural networks (CNNs) have dominated the field, achieving significant success in 3D medical image segmentation. However, CNNs struggle with capturing long-range dependencies and global context, limiting their performance, particularly for fine and complex structures. Recent transformer-based models, such as TransUNet and nnFormer, have demonstrated promise in addressing these limitations, though they still rely on hybrid CNN-transformer architectures. This paper introduces a novel, fully convolutional-free model based on transformer architecture and self-attention mechanisms for 3D medical image segmentation. Our approach focuses on improving multi-semantic segmentation accuracy and addressing domain adaptation challenges between thick and thin slice CT images. We propose a joint loss function that facilitates effective segmentation of thin slices based on thick slice annotations, overcoming limitations in dataset availability. Furthermore, we present a benchmark dataset for multi-semantic segmentation on thin slices, addressing a gap in current medical imaging research. Our experiments demonstrate the superiority of the proposed model over traditional and hybrid architectures, offering new insights into the future of convolution-free medical image segmentation.