•  
  •  
 

Abstract

Lung nodule cancer detection is a critical and complex medical challenge. Accuracy in detecting lung nodules can significantly improve patient prognosis and care. The main challenge is to develop a detection method that can accurately distinguish between benign and malignant nodules and perform effectively under various imaging conditions. The development of technology and investment in deep learning techniques in the medical field make it easy to use Positron Emission Tomography (PET) and Computed Tomography (CT). Thus, this paper presents lung cancer detection by filtering the PET-CT image, obtaining the lung region of interest (ROI), and training using Convolution neural network (CNN)-Deep learning models for defending the nodules’ location. The limitation dataset composed of 220 cases with 560 nodules with fixed Hounsfield Units (HU) is used to increase the training’s speed and save data. The trained models involve CNN, DCNN, 3DCNN, VGG 19, ResNet 18, Inception V1, and Inception-ResNet to detect the lung nodules. The experiment shows high-speed training with VGG 19 outperforming the rest of deep learning, it achieves accuracy, Precision, Specificity, Sensitivity, F1-Score, IoU, FP rate with standard division; 98.65 ± 0.22, 98.80 ± 0.15, 98.70 ± 0.20, 98.55 ± 0.18, 98.60 ± 0.16, 0.94 ± 0.03, 1.05 ± 0.22, respectively. Moreover, the experiment results show an overall error rate and a standard division between ± 0.04 to ± 0.54 distributed over the calculation terms.

Keywords

Convolution neural network (CNN), CT image, Deep learning; Lung cancer, Lung nodules

Subject Area

Physics

Article Type

Article

First Page

1596

Last Page

1608

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Download

Share

COinS