A lightweight CNN for enhanced non-small cell lung cancer classification using CT scan image

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer is a major cause of cancer mortality globally.
• Accurate and early detection of lung cancer is crucial for patient survival.
• Radiologist expertise impacts diagnostic accuracy in lung cancer detection using CT scans.
• Deep learning offers potential for consistent and reliable clinical decision support in cancer diagnosis.

Purpose of the Study:

• To propose MiniConvNet, a lightweight convolutional neural network for detecting and classifying non-small cell lung cancer (NSCLC) and its subtypes from CT images.
• To evaluate the generalizability of MiniConvNet on histopathological lung cancer data.
• To benchmark MiniConvNet against established CNN architectures.

Main Methods:

• Development of MiniConvNet, a lightweight CNN architecture.
• Training and evaluation on two public lung cancer datasets (CT and histopathology).
• Benchmarking against ResNet50, VGG16, VGG19, Inception V3, MobileNetV3Small, EfficientNetV2B0, and ConvNeXtTiny.

Main Results:

• MiniConvNet achieved competitive or superior performance compared to baseline models.
• The model demonstrated robustness across different imaging modalities (CT and histopathology).
• MiniConvNet has a significantly smaller model size and faster inference speed than established architectures.

Conclusions:

• MiniConvNet is a promising, efficient tool for lung cancer subtype classification.
• The model's lightweight nature and speed make it suitable for resource-constrained clinical environments.
• MiniConvNet demonstrates potential for improving lung cancer diagnosis accuracy and efficiency.