Abstract:
Lung cancer is the leading cause of cancer death in the globe. Early identification of
lung cancer can help to prevent lung cancer from becoming chronic, severe and life
threatening. Here, CT images are frequently used and an automated and appropriate
method using deep learning technique that can potentially makes a huge contribute to
make quick and accurate diagnosis for lung cancer. However, in the area of medical
imaging using deep learnings techniques, there have two limitations. One huge training
time and the other one is insufficient and imbalanced datasets. This study will present
the number of image balancing and reducing the overall processing time. The dataset
we use in this research work contains three cases normal, benign, and malignant. In the
dataset, we use data augmentation techniques to increase the amount of data then we
apply some image processing method on the dataset including some filter like
GaussianBlur for reduce noise, Adaptive Thresholding for high component details and
edge, lastly Image Negative and Bit Plane Slicing. We proposed a Customized
Convolutional Neural Network (CCNN) model using 224 x 224 size images classify
the lung cancer into three classes. Seven transfer learning models, VGG16, VGG19,
ResNet50, ResNet101, DenseNet201, EfficientNetB4 and MobileNetV2 are applied
with the same image and batch size and all the transfer leaning models are compared
with the proposed CCNN model and we got the maximum test accuracy of 98.18%
from CCNN model including the require time for per epoch 2 sec. Our proposed model
may help medical expert for diagnosis the lung cancer from medical CT images and our
aim to add more data in the dataset and use more deep learning models in our future
study.
