基于MedSAM模型的视网膜血管图像分割

打开文本图片集

中图分类号：TP391.4；TP18 文献标识码：A 文章编号：2096-4706（2025）11-0054-05

Retinal Vessel Image Segmentation Based on the MedSAM Model

LIUYahui

(ZhejiangChineseMedicalUniversity,Hangzhou31oo53,China)

Abstract:Thesegmentationofretinalblood vessls in fundus images isof great significance for theearlydetectionand diagnosisofoculardiseases.Traditionalautomaticretinalsegmentation methodsrelyonlargeamountsoflabeleddata formodel pre-training.However,inpracticalaplications,ecaleofretialimagedatasetsislmited,andlabelingcostsarehghThis studyappliesageneralmedicalimage segmentationmodel,the Medical Segment Anything Model (MedSAM),fortransfer leaming toachieve acurateretinal segmentation under conditions of limiteddatasets.Thisstudy isevaluatedonmainstream retinal fundus imagedatasets,CHASE_DBlandSTARE.Theexperimentalresults indicate thatcompared withU-Netand U-Net ++ models,MedSAM's losscanconverge to a stable interval faster during training,and its segmentation performance leads inthreemetricsofDice Sore,IU,andPA,demonstrating excellentsegmentationpeformanceadprovidingapowerfultolfor clinical diagnosis.

Keywords:MedSAM; retina; image segmentation; medical image

0 引言

视网膜包含大量血管，是体内唯一可以通过非侵入性手段深入观察的血管系统[]。（剩余12803字）