The School of Medical Information Engineering, Anhui University of Chinese Medicine, Hefei, China.
NHC Key Laboratory of Hormones and Development, Chu Hsien-I Memorial Hospital and Tianjin Institute of Endocrinology, Tianjin Medical University, Tianjin, China.
PLoS One. 2023 May 5;18(5):e0282350. doi: 10.1371/journal.pone.0282350. eCollection 2023.
Breast cancer is a major health problem with high mortality rates. Early detection of breast cancer will promote treatment. A technology that determines whether a tumor is benign desirable. This article introduces a new method in which deep learning is used to classify breast cancer.
A new computer-aided detection (CAD) system is presented to classify benign and malignant masses in breast tumor cell samples. In the CAD system, (1) for the pathological data of unbalanced tumors, the training results are biased towards the side with the larger number of samples. This paper uses a Conditional Deep Convolution Generative Adversarial Network (CDCGAN) method to generate small samples by orientation data set to solve the imbalance problem of collected data. (2) For the high-dimensional data redundancy problem, this paper proposes an integrated dimension reduction convolutional neural network (IDRCNN) model, which solves the high-dimensional data dimension reduction problem of breast cancer and extracts effective features. The subsequent classifier found that by using the IDRCNN model proposed in this paper, the accuracy of the model was improved.
Experimental results show that IDRCNN combined with the model of CDCGAN model has superior classification performance than existing methods, as revealed by sensitivity, area under the curve (AUC), ROC curve and accuracy, recall, sensitivity, specificity, precision,PPV,NPV and f-values analysis.
This paper proposes a Conditional Deep Convolution Generative Adversarial Network (CDCGAN) which can solve the imbalance problem of manually collected data by directionally generating small sample data sets. And an integrated dimension reduction convolutional neural network (IDRCNN) model, which solves the high-dimensional data dimension reduction problem of breast cancer and extracts effective features.
乳腺癌是一种死亡率较高的重大健康问题。早期发现乳腺癌将促进治疗。一种能够确定肿瘤是良性的技术是可取的。本文介绍了一种利用深度学习对乳腺癌进行分类的新方法。
提出了一种新的计算机辅助检测（CAD）系统，用于对乳腺肿瘤细胞样本中的良性和恶性肿块进行分类。在 CAD 系统中，（1）对于肿瘤数量不平衡的病理数据，训练结果偏向于样本数量较大的一侧。本文使用条件深度卷积生成对抗网络（CDCGAN）方法通过定向数据集生成小样本，以解决所采集数据的不平衡问题。（2）针对高维数据冗余问题，本文提出了一种集成降维卷积神经网络（IDRCNN）模型，解决了乳腺癌高维数据降维和提取有效特征的问题。随后的分类器发现，通过使用本文提出的 IDRCNN 模型，模型的准确性得到了提高。
实验结果表明，与现有方法相比，IDRCNN 与 CDCGAN 模型相结合具有优越的分类性能，通过灵敏度、曲线下面积（AUC）、ROC 曲线和准确性、召回率、灵敏度、特异性、精度、PPV、NPV 和 f 值分析得到证实。
本文提出了一种条件深度卷积生成对抗网络（CDCGAN），可以通过定向生成小样本数据集来解决手动收集数据的不平衡问题。并提出了一种集成降维卷积神经网络（IDRCNN）模型，解决了乳腺癌高维数据降维和提取有效特征的问题。
