基于深度学习的图像智能识别算法在食管癌病理诊断应用的初步研究OA
Preliminary Study on the Application of Image Intelligent Recognition Algorithm Based on Deep Learning in Pathological Diagnosis of Esophageal Cancer
目的 构建一套高效的食管 HE 组织切片图像智能识别算法,以辅助医师提升诊断效率与准确性.方法 基于人工智能,采用全视野扫描图像下的食管癌 HE 组织切片进行模型训练,并以卷积神经网络(CNN)为基础,构建两种自动识别框架.其一为基础 CNN 框架,以 ResNet-18 为骨干网络,用于保障基础分类能力;其二为针对小样本分类性能与特征区分难题,提出 ACNNC(Attention+CNN+CLAHE)框架.该框架将对比度限制自适应直方图均衡化(CLAHE)预处理与通道—空间双注意力机制深度融合,以解决食管癌 HE 切片中细微病变特征提取难的问题.结果 ACNNC 对食管癌 HE 组织切片影像的识别性能稳定,F1 分数达到 0.80,召回率为 0.81,精确率为 0.75;与现有主流模型相比,本文模型在各指标上均显著更优(P<0.05).结论 本研究为计算机辅助食管癌病理诊断提供了解决方案,有助于改善食管癌病理诊断中重复性低、漏诊、误诊等问题,具备重要临床应用价值.
Objective To develop an efficient intelligent recognition algorithm for esophageal HE tissue section images to assist physicians in improving diagnostic efficiency and accuracy.Methods Based on artificial intelligence,whole-slide images of hematoxylin and eosin(HE)-stained tissue sections from esophageal cancer were used for model training.Two automatic recognition frameworks were constructed on the basis of convolutional neural networks(CNN):one was a basic CNN framework with ResNet-18 as the backbone to ensure basic classification capability;The second was the ACNNC(Attention+CNN+CLAHE)framework,proposed to address the challenges of classification performance and feature discrimination in small sample settings.It integrates Contrast Limited Adaptive Histogram Equalization(CLAHE)preprocessing with a channel-spatial dual attention mechanism to tackle the difficulty in extracting subtle lesion features from HE slides of esophageal cancer.Results The ACNNC framework showed stable recognition performance for esophageal carcinoma HE-stained tissue section images,with an F1-score of 0.80,recall of 0.81,and precision of 0.76.Compared with existing mainstream models,all indicators were significantly superior to those of the control models(P<0.05).Conclusion This study provides a solution for computer-aided pathological diagnosis of esophageal carcinoma,helping to address issues such as low repeatability,missed diagnosis,and misdiagnosis in esophageal carcinoma diagnosis,and has important clinical application value.
王富强;王伟;原玉芬;孟翠丽;徐楠;杨海军
河南科技大学附属安阳市肿瘤医院,河南 安阳,455000河南省安阳市工学院计算机学院,河南 安阳,455000河南科技大学附属安阳市肿瘤医院,河南 安阳,455000河南科技大学附属安阳市肿瘤医院,河南 安阳,455000河南科技大学附属安阳市肿瘤医院,河南 安阳,455000河南科技大学附属安阳市肿瘤医院,河南 安阳,455000
医药卫生
深度学习食管癌目标检测注意力机制计算机辅助诊断
deep learningesophageal cancerobject detectionattention mechanismcomputer aided diagnosis
《食管疾病》 2026 (3)
180-184,5
河南省科技攻关计划项目(232102310222)
评论