基于深度学习的H&E染色图像在结直肠癌中的微生物预测研究  

Research on Microorganism Prediction in H&E Stained Images Based on Deep Learning

作　　者：张艳萍 胡丰源 韩佳颖 黄雅萍 王海燕 张洋洋 赵雪柠 刘春明 ZHANG Yan-ping;HU Feng-yuan;HAN Jia-ying;HUANG Ya-ping;WANG Hai-yan;ZHANG Yang-yang;ZHAO Xue-ning;LIU Chun-ming(College of Mathematical Sciences and Engineering,Hebei University of Engineering,Handan,Hebei,056107,China)

机构地区：[1]河北工程大学数理科学与工程学院,河北邯郸056107

出　　处：《现代生物医学进展》2025年第5期822-832,共11页Progress in Modern Biomedicine

摘　　要：目的:本研究旨在利用深度学习技术分析结直肠癌(CRC)病理切片图像,预测与结直肠癌相关的微生物丰度。方法:研究团队整合了TCGA数据库中的病理图像与微生物数据,开发了MDLR-Mean(Microbe Deep Learning Regression Prediction model Mean)模型。该模型融合了TOAD(Tumour Origin Assessment via Deep Learning)的肿瘤起源评估能力和MLP的深度学习特性,通过特征聚合技术提升预测精度,并采用MAE损失函数优化模型表现。结果:实验结果显示,MDLR-Mean模型在微生物丰度预测上表现卓越,在皮尔逊相关系数(PCC)、均方误差(MSE)和平均绝对误差(MAE)评估指标上均表现优异(P<0.05)。尤其是平均PCC相较于现有方法提升了36.5%,验证了模型的高效性和准确性。结论:本研究成功验证了MDLR-Mean模型在预测结直肠癌病理切片图像中微生物丰度方面的高准确性和可靠性,揭示深度学习将在未来结直肠癌诊治中发挥重要作用和助力精准医疗。Objective:The aim of this study was to analyse colorectal cancer(CRC)pathology slice im ages using deep learning techniques to predict microbial abundance associated with colorectal cancer.Methods:The research team integrated pathological images and microbiological data from the TCGA database to develop the MDLR-Mean(Microbe Deep Learning Regression Prediction model Mean)model.The model integrates the tumour origin assessment capability of TOAD(Tumour Origin Assessment via Deep Learning)and the deep learning features of MLP,improves the prediction accuracy by feature aggregation technology,and optimizes the model performance by using MAE loss function.Results:The experimental results showed that the MDLR-Mean model performed excellently in microbial abundance prediction,and excelled in PCC,MSE,and MAE assessment indexes(P<0.05).Especially,the average PCC was improved by 36.5%compared with the existing methods,which verified the efficiency and accuracy of the model.Conclusion:This study successfully validated the high accuracy and reliability of the MDLR-Mean model in predicting microbial abundance in colorectal cancer pathological section images,suggesting that deep learning will play an important role in the future diagnosis and treatment of colorectal cancer and help precision medicine.

关 键 词：结直肠癌 病理切片图像 深度学习 微生物丰度预测 MDLR-Mean模型 

分 类 号：R3[医药卫生—基础医学] Q939.99[生物学—微生物学] R735.35