基于深度学习的非小细胞肺癌检测  

Detection of Non-Small Cell Lung Cancer Based on Deep Learning

作　　者：贺晓松 胡川丽 赵江 HE Xiaosong;HU Chuanli;ZHAO Jiang(School of Big Data and Artificial Intelligence,Chongqing Institute of Engineering,Chongqing 400056,P.R.China)

机构地区：[1]重庆工程学院大数据与人工智能学院,400056 [2]陆军军医大学第二附属医院麻醉科,重庆400037 [3]陆军军医大学第二附属医院泌尿科,重庆400037

出　　处：《临床放射学杂志》2025年第2期265-273,共9页Journal of Clinical Radiology

摘　　要：目的为实现高效和准确地使用基于胸部CT影像的非小细胞肺癌检测,而提出一种基于EfficientNetB0-CBAM模型的深度学习算法。方法搜集包含4种类型的1000张CT影像,构建肺癌图像分类数据集。选择EfficientNetB0作为基础分类模型,将CBAM注意力模块集成到EfficientNetB0中构建EfficientNetB0-CBAM模型,增强重要特征并抑制无效特征,同时使用批归一化和随机失活模块加速网络的训练并减轻过拟合程度。以准确率、精准率、召回率、F1-Score作为性能评估指标,在测试集上测试模型性能,并与其他SOTA模型(VGG16、ResNet50、MobileNet、DenseNet121、ConvNeXtTiny)进行比较。结果EfficientNetB0-CBAM模型实现了95.56%的准确率、95.49%的精准率、96.13%的召回率和95.74%的F1-Score,高于对比实验中使用的其他深度学习方法。使用Grad-CAM可视化方法表明EfficientNetB0-CBAM模型可以更多地关注和识别起关键作用的病灶区域。结论本研究为深度学习技术在非小细胞肺癌检测的自动识别检测提供技术支撑,为肺癌的早期诊断提供依据。Objective To achieve efficient and accurate detection of non-small cell lung cancer(NSCLC)based on chest CT images,a deep learning algorithm based on the EfficientNetB0-CBAM model is proposed.Methods A dataset of lung cancer image classification was constructed using 1000 CT images of four types.EfficientNetB0 was chosen as the base classification model,and the CBAM attention mechanism was integrated into EfficientNetB0 to build the EfficientNetB0-CBAM model,which enhances important features and suppresses irrelevant ones.Batch normalization and random dropout modules were used to accelerate network training and alleviate overfitting.Accuracy,precision,recall,and F1-Score were used as perform ance evaluation metrics to test the model's performance on the test set,and the results were compared with other state-of-the-art(SOTA)models(VGG16,ResNet50,MobileNet,DenseNet121,ConvNeXtTiny).Results The EfficientNetB0-CBAM model achieved 95.56%accuracy,95.49%precision,96.13%recall,and 95.74%F1-Score,outperforming other deep learning methods used in the comparison experiments.The Grad-CAM visualization method showed that the EfficientNetB0-CBAM model focuses more on lesion areas that play a key role in recognition.Conclusion This study provides technical support for the automatic detection of NSCLC using deep learning and offers a basis for the early diagnosis of lung cancer.

关 键 词：肺癌检测 深度学习 注意力机制 非小细胞肺癌 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程] R734.2[医药卫生—肿瘤]