Lung Cancer Classification Using Modified Squeezenet

Abstract

Lung cancer is the primary cause of mortality in individuals diagnosed with cancer. Detecting and diagnosing lung cancer early significantly reduces the mortality rate. The early diagnosis of lung cancer is greatly facilitated by medical imaging. The recommendation is to undergo a CT scan, as it has a higher probability of detecting lung cancer during its initial phases. The detection of lung cancer greatly depends on the utilization of advanced deep learning technology, specifically convolutional neural networks, which assist in accurately classifying the CT image. This paper proposed a Modified light weight SqueezeNet architecture that mixes bottleneck residual network and fully connected layer along with global average pooling in the original network. This modification enhances the classification performance with a slight rise in computational complexity.  CT images of 330 patients are used as a data set for testing the proposed technique, which is executed in MATLAB 2022a platform. The proposed method can identify lung cancer and categorize it as either malignant or normal with test Accuracy of 95.76%, Recall-92.94%, Precision of 98.75%, Specificity-98.75%, and AUC-0.9977. The Modified SqueezeNet gives better classification performance against the base SqueezeNet model. The proposed method outperforms traditional deep learning networks like AlexNet, ShuffleNet, ResNet-50, and GoogleNet.