Classifiers Combined with DenseNet Models for Lung Cancer Computed Tomography Image Classification: A Comparative...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer is a leading cause of mortality globally.
• Deep learning shows potential in medical image analysis.
• Limited comparative studies exist on classifier combinations with DenseNet for lung cancer classification.

Purpose of the Study:

• To compare the performance of different classifier combinations (SVM, ANN, MLP) with DenseNet architectures for lung cancer classification.
• To evaluate the effectiveness of these combinations using chest CT scan images.
• To establish benchmarks for automated lung cancer detection systems.

Main Methods:

• A comparative analysis of 1,000 chest CT scans (Adenocarcinoma, Large Cell Carcinoma, Squamous Cell Carcinoma, normal).
• Three DenseNet variants (DenseNet-121, DenseNet-169, DenseNet-201) combined with SVM, ANN, and MLP classifiers.
• Performance evaluated using accuracy, AUC, precision, recall, specificity, and F1-score with an 80-20 train-test split.

Main Results:

• Optimal model achieved 92% training accuracy and 83% test accuracy.
• Performance across models ranged from 73% to 83% test accuracy.
• The MLP-DenseNet-169 combination showed 83% test accuracy, with SVM-DenseNet-169 demonstrating superior stability.

Conclusions:

• Deep learning effectively categorizes chest CT scans for lung cancer detection.
• The MLP-DenseNet-169 combination provides a promising benchmark for automated lung cancer detection.
• Further research should address limitations such as retrospective design and sample size.