Institute of Information Technology, Jahangirnagar University, Savar, Dhaka, 1342, Bangladesh.
Institute of Information Technology, Jahangirnagar University, Savar, Dhaka, 1342, Bangladesh; Queensland University of Technology, 2 George St, Brisbane City, QLD, 4000, Australia.
Comput Biol Med. 2021 Dec;139:104961. doi: 10.1016/j.compbiomed.2021.104961. Epub 2021 Oct 27.
Lung cancer, also known as pulmonary cancer, is one of the deadliest cancers, but yet curable if detected at the early stage. At present, the ambiguous features of the lung cancer nodule make the computer-aided automatic diagnosis a challenging task. To alleviate this, we present LungNet, a novel hybrid deep-convolutional neural network-based model, trained with CT scan and wearable sensor-based medical IoT (MIoT) data. LungNet consists of a unique 22-layers Convolutional Neural Network (CNN), which combines latent features that are learned from CT scan images and MIoT data to enhance the diagnostic accuracy of the system. Operated from a centralized server, the network has been trained with a balanced dataset having 525,000 images that can classify lung cancer into five classes with high accuracy (96.81%) and low false positive rate (3.35%), outperforming similar CNN-based classifiers. Moreover, it classifies the stage-1 and stage-2 lung cancers into 1A, 1B, 2A and 2B sub-classes with 91.6% accuracy and false positive rate of 7.25%. High predictive capability accompanied with sub-stage classification renders LungNet as a promising prospect in developing CNN-based automatic lung cancer diagnosis systems.
肺癌，又称支气管癌，是最致命的癌症之一，但如果在早期发现，是可以治愈的。目前，肺癌结节的模糊特征使得计算机辅助自动诊断成为一项具有挑战性的任务。为了缓解这一问题，我们提出了 LungNet，这是一种基于新型混合深度卷积神经网络的模型，使用 CT 扫描和可穿戴传感器的医疗物联网（MIoT）数据进行训练。LungNet 由一个独特的 22 层卷积神经网络（CNN）组成，它结合了从 CT 扫描图像和 MIoT 数据中学习到的潜在特征，以提高系统的诊断准确性。该网络在集中式服务器上运行，使用一个平衡的数据集进行训练，该数据集包含 525000 张图像，能够以高精度（96.81%）和低假阳性率（3.35%）将肺癌分为五类，优于类似的基于 CNN 的分类器。此外，它还能以 91.6%的准确率和 7.25%的假阳性率将 1A、1B、2A 和 2B 期肺癌分为亚类。高预测能力和亚阶段分类使 LungNet 成为开发基于 CNN 的自动肺癌诊断系统的一个有前途的选择。
