基于Mamba改进的3D肝脏及肝肿瘤CT图像分割

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2026）03-0082-06

Improved 3D Liver and Liver Tumor CT Image Segmentation Based on Mamba

GUOJiahao，HUHuaifei （South-Central Minzu University，Wuhan 43oo74， China）

Abstract： Accurate segmentation of liver and liver tumors in three-dimensional Computed Tomography （CT） images is crucialforclinicaldiagnosisandtreatmentplanning.ToaddresstheissuesthatexistingConvolutionalNeuralNetwork（CNN） segmentationmethodsstrugle tocapturelong-rangedependenciesduetolimitedreceptivefelds，andTransformersegmentation methodsarerestricted inapplicationonthree-dimensional medicalimages with limitedsamples，this paper proposesanew 3D medicalimage segmentation model based on Mamba.The model contains a 3D gated spatial convolution module andachannel shuffle upsampling module， which extract spatial correlation features and common features，respectively.The two modules synergistically enhance themodeling abilityofthe modelfor global tructures，while simultaneouslyensuring precisebondary positioning.ExperimentalresultsontheLiTS2o17datasetshow thatthe Dice SimilarityCoeffcient （DSC）of liversegmentation reaches 96.42% ，and theDSC of liver tumor segmentation reaches 70.70% .In the generalization experiment on the 3D-IRCADb dataset，the DSCsof liver and liver tumor segmentationreach 96.79% and 67.10% ，respectively.Resultsof multiple comparative experiments further verify the superiority and robustness of the proposed model in segmentation performance.

Keywords： liver tumor segmentation; Mamba; State Space Model; Deep Learning

0 引言

医学图像分割在疾病诊断和治疗规划中发挥着重要作用，这一特点在肝肿瘤治疗领域尤为显著[1-3]精准的分割结果将直接影响手术方案的设计与术后评估。（剩余11732字）