Deep Learning-Based Cancerous Lung Nodule Detection in Computed Tomography Imageries


Kumar S. V., Chen F., Kim S., Choi J.

4th International Conference on Intelligent and Fuzzy Systems (INFUS), Bornova, Turkey, 19 - 21 July 2022, vol.505, pp.44-52 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Volume: 505
  • Doi Number: 10.1007/978-3-031-09176-6_5
  • City: Bornova
  • Country: Turkey
  • Page Numbers: pp.44-52
  • Keywords: Deep learning, Computed tomography, Lung cancer, Nodules, Malignant detection, CNN, U-Net model, LUNA-16
  • Istanbul Technical University Affiliated: No

Abstract

Computed tomography images have been widely used for lung cancer diagnostics. In the early stages of lung cancers, lung nodules are tiny, and even radiologists struggle to detect and diagnose them. Conventional and manual detection of lung nodules is a sequential and time-consuming process for radiologists. On the other hand, deep-learning-based automatic detection algorithm is an alternative approach that has recently drawn much attention. The deep learning method becomes successful and outperforms physicians in classification, nodule deduction, and false-positive reduction of malignant pulmonary nodules on chest radiograph. Accurate detection of early lung cancer nodules can be critical and hence improve the cure rate of lung cancer. In this paper, a novel deep learning-based lung nodule detection method is presented for automatic detection of malignant tumors in the lungs. The proposed 3-D CNN model classifies the candidates as nodules or non-nodules, while 2-D U-Net is used to segment the position of lung nodules. Here, the data augmentation technique is used to generate a large number of training examples, and regularization is also applied to avoid overfitting. The evaluation metrics adopted in this paper are the dice coefficient loss and the area under the receiver operating characteristic curve, which are frequently used in image segmentation tasks. The performance of the proposed method has been verified by using LUNA-16, which is a publicly available medical dataset. The simulation results show that the proposed method can achieve superior detection accuracy, which surpasses the conventional methods.