半监督知识蒸馏的组织病理分类模型

打开文本图片集

关键词：组织病理分类；先验知识；自监督对比学习；知识蒸馏；半监督学习DOI：10.15938/j. jhust. 2025.05.001中图分类号：TP315.69 文献标志码：A 文章编号：1007-2683（2025）05-0001-15

Histopathological Classification Model for Semi-supervised Knowledge Distillation

XIE Yining'，CHEN Xiaokai²， ZHANG Yuming²， ZHU Yinping²， LONG Jun1 （1. Enginering， ，15O040，China; 2. Computer Control ， ， 15O040，China）

Abstract：Histopathologicalimageclasificationisakeytechnologyforcomputer-aideddiagnosis，whichlargelyreliesonalarge amount labeleddata.Although boostingsemi-supervised weaklysupervised methodscanreduce dependenceonlabeled data， theystillsufffrosussuchslodatautiztioxcsveodelparamesdpoodelgneralaiabilityodre thesechalenges，thispaperproposesasemi-supervisedlearningframework（BSL）basedonknowledgedistilltion.Thismethodfist utilieself-supeisedtrastivelaingtprainaistopaologialiagesificationodeltatitegatesprioroldgef cellnuclei，asthe teachermodel.Then，theteacher modelisfurthertrainedusingenhancedsemi-supervisedlearingmethods. Finally，theteachermodelisusedtoguidethetrainingthestudentmodelachieving knowledgetransfer，aceleratingmodel convergence，improvingitsperformance.Theexperimentalresultsshowthattheproposedmethodoutperforms existing methodson public datasets.

Keywords：histopathological clasification；priorknowledge；self-supervisedcontrastivelearing；knowledgedistilation；semi supervised learning

0 引言

胃癌是一种常见的恶性肿瘤，其发病率和死亡率在全球范围内居高不下，尤其在一些发展中国家和地区尤为突出。（剩余21755字）