Department of BME, Alpha College of Engineering, Chennai-124, India.
Department of ECE, St. Peter's Institute of Higher Education and Research, Avadi, Chennai-54, India.
Med Biol Eng Comput. 2022 Jan;60(1):221-228. doi: 10.1007/s11517-021-02462-3. Epub 2021 Nov 22.
The early detection of pulmonary nodules using computer-aided diagnosis (CAD) systems is very essential in reducing mortality rates of lung cancer. In this paper, we propose a new deep learning approach to improve the classification accuracy of pulmonary nodules in computed tomography (CT) images. Our proposed CNN-5CL (convolutional neural network with 5 convolutional layers) approach uses an 11-layer convolutional neural network (with 5 convolutional layers) for automatic feature extraction and classification. The proposed method is evaluated using LIDC/IDRI images. The proposed method is implemented in the Python platform, and the performance is evaluated with metrics such as accuracy, sensitivity, specificity, and receiver operating characteristics (ROC). The results show that the proposed method achieves accuracy, sensitivity, specificity, and area under the roc curve (AUC) of 98.88%, 99.62%, 93.73%, and 0.928, respectively. The proposed approach outperforms various other methods such as Naïve Bayes, K-nearest neighbor, support vector machine, adaptive neuro fuzzy inference system methods, and also other deep learning-based approaches.
利用计算机辅助诊断（CAD）系统早期发现肺结节对于降低肺癌死亡率非常重要。在本文中，我们提出了一种新的深度学习方法来提高计算机断层扫描（CT）图像中肺结节的分类准确性。我们提出的 CNN-5CL（具有 5 个卷积层的卷积神经网络）方法使用 11 层卷积神经网络（具有 5 个卷积层）进行自动特征提取和分类。使用 LIDC/IDRI 图像评估所提出的方法。所提出的方法在 Python 平台上实现，并使用准确性、敏感性、特异性和接收器操作特征（ROC）等指标来评估性能。结果表明，所提出的方法分别达到了 98.88%、99.62%、93.73%和 0.928 的准确性、敏感性、特异性和 ROC 下面积。所提出的方法优于贝叶斯、K-最近邻、支持向量机、自适应神经模糊推理系统方法等各种其他方法，也优于其他基于深度学习的方法。
