College of Information Science and Engineering, Xinjiang University, Urumqi 830046, China.
Department of Gynecology, The First Affiliated Hospital of Xinjiang Medical University, Urumqi, PR China.
Photodiagnosis Photodyn Ther. 2023 Jun;42:103557. doi: 10.1016/j.pdpdt.2023.103557. Epub 2023 Apr 13.
Cervical cancer is the most common reproductive malignancy in the female reproductive system. The incidence rate and mortality rate of cervical cancer among women in China are high. In this study, Raman spectroscopy was used to collect tissue sample data from patients with cervicitis, cervical precancerous low-grade lesions, cervical precancerous high-grade lesions, well differentiated squamous cell carcinoma, moderately differentiated squamous cell carcinoma, poorly differentiated squamous cell carcinoma and cervical adenocarcinoma. The collected data were preprocessed using an adaptive iterative reweighted penalized least squares (airPLS) algorithm and derivatives. Convolutional neural network (CNN) and residual neural network (ResNet) classification models were constructed to classify and identify seven types of tissue samples. The attention mechanism efficient channel attention network (ECANet) module and squeeze-and-excitation network (SENet) module were combined with the established CNN and ResNet network models, respectively, to make the models have better diagnostic performance. The results showed that efficient channel attention convolutional neural network (ECACNN) had the best discrimination, and the average accuracy, recall, F1 and AUC values after five cross-validations could reach 94.04%, 94.87%, 94.43% and 96.86%, respectively.
宫颈癌是女性生殖系统中最常见的生殖系统恶性肿瘤。中国女性宫颈癌的发病率和死亡率都很高。本研究采用激光共聚焦拉曼光谱技术采集宫颈炎、宫颈低级别上皮内瘤变、宫颈高级别上皮内瘤变、高分化鳞癌、中分化鳞癌、低分化鳞癌和宫颈腺癌患者的组织样本数据。采集的数据采用自适应迭代重加权惩罚最小二乘（airPLS）算法和导数进行预处理。构建卷积神经网络（CNN）和残差神经网络（ResNet）分类模型，对七种组织样本进行分类和识别。分别将注意力机制高效通道注意力网络（ECANet）模块和压缩激励网络（SENet）模块与建立的 CNN 和 ResNet 网络模型相结合，使模型具有更好的诊断性能。结果表明，高效通道注意力卷积神经网络（ECACNN）的鉴别能力最好，五次交叉验证后的平均准确率、召回率、F1 值和 AUC 值分别达到 94.04%、94.87%、94.43%和 96.86%。
