[Paper Review] LCDctCNN: Lung Cancer Diagnosis of CT scan Images Using CNN Based Model

Muntasir Mamun, Md Ishtyaq Mahmud+2 more|arXiv (Cornell University)|Apr 10, 2023

Lung Cancer Diagnosis and TreatmentMedicine3 citations

TL;DR

This paper proposes LCDctCNN, a custom Convolutional Neural Network (CNN) model for early lung cancer diagnosis using CT scan images. The model achieves 92% accuracy, 98.21% AUC, 91.72% recall, and 0.328 loss, outperforming benchmark models like Inception V3, Xception, and ResNet-50 in diagnostic performance on a CT imaging dataset.

ABSTRACT

The most deadly and life-threatening disease in the world is lung cancer. Though early diagnosis and accurate treatment are necessary for lowering the lung cancer mortality rate. A computerized tomography (CT) scan-based image is one of the most effective imaging techniques for lung cancer detection using deep learning models. In this article, we proposed a deep learning model-based Convolutional Neural Network (CNN) framework for the early detection of lung cancer using CT scan images. We also have analyzed other models for instance Inception V3, Xception, and ResNet-50 models to compare with our proposed model. We compared our models with each other considering the metrics of accuracy, Area Under Curve (AUC), recall, and loss. After evaluating the model's performance, we observed that CNN outperformed other models and has been shown to be promising compared to traditional methods. It achieved an accuracy of 92%, AUC of 98.21%, recall of 91.72%, and loss of 0.328.

Motivation & Objective

• To develop a deep learning model for early and accurate lung cancer diagnosis using CT scan images.
• To improve upon traditional diagnostic methods by leveraging the power of convolutional neural networks in medical imaging.
• To compare the performance of the proposed CNN model against established architectures like Inception V3, Xception, and ResNet-50.
• To evaluate model effectiveness using standard metrics including accuracy, AUC, recall, and loss.
• To contribute a robust, high-precision diagnostic tool for clinical use in lung cancer screening.

Proposed method

• A custom CNN architecture, named LCDctCNN, was designed specifically for classifying lung cancer from CT scan images.
• The model was trained and validated on a dataset of CT scan images, using standard data augmentation techniques to improve generalization.
• Performance was evaluated using key deep learning metrics: accuracy, area under the ROC curve (AUC), recall, and loss.
• The model was compared against transfer learning-based models: Inception V3, Xception, and ResNet-50, all fine-tuned on the same dataset.
• Training utilized stochastic gradient descent with cross-entropy loss to optimize classification performance.
• The architecture leverages multiple convolutional and pooling layers to extract hierarchical features from lung CT images.

Experimental results

Research questions

• RQ1 Can a custom CNN model achieve superior diagnostic performance compared to established pre-trained models in lung cancer detection from CT scans?
• RQ2 What is the diagnostic accuracy, AUC, recall, and loss of the proposed LCDctCNN model on a CT imaging dataset?
• RQ3 How does the performance of LCDctCNN compare to Inception V3, Xception, and ResNet-50 in terms of key evaluation metrics?
• RQ4 Does the proposed CNN model demonstrate improved generalization and robustness for early lung cancer detection?
• RQ5 Can deep learning models significantly reduce false negatives in lung cancer screening using CT imaging?

Key findings

• The LCDctCNN model achieved 92% classification accuracy on the lung cancer CT scan dataset.
• The model recorded an AUC of 98.21%, indicating excellent discrimination between cancer and non-cancer cases.
• Recall was measured at 91.72%, suggesting strong sensitivity in detecting positive lung cancer cases.
• The model's loss value was 0.328, indicating effective optimization and convergence during training.
• LCDctCNN outperformed Inception V3, Xception, and ResNet-50 across all evaluation metrics, demonstrating superior performance.
• The results confirm that a custom CNN architecture can be more effective than transfer learning models for this specific diagnostic task.