Lung Cancer Diagnosis From Computed Tomography Images Using Deep Learning Algorithms With Random Pixel Swap Data...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Deep learning (DL) shows potential for automated lung cancer diagnosis from CT scans.
• Limited clinical data and diverse DL architectures pose challenges for diagnostic performance.
• Existing data augmentation (DA) methods often struggle with chest CT scan data.

Purpose of the Study:

• To introduce Random Pixel Swap (RPS), a novel DA technique for enhancing DL models in lung cancer diagnosis.
• To evaluate RPS's effectiveness on convolutional neural networks (CNNs) and transformers using CT scan images.
• To improve the accuracy and reliability of automated lung cancer detection.

Main Methods:

• RPS generates augmented data by randomly swapping pixels within CT scan images.
• Evaluated RPS on ResNet, MobileNet, Vision Transformer, and Swin Transformer models.
• Utilized two public CT datasets and measured accuracy and AUROC, with statistical significance assessed via paired t tests.

Main Results:

• RPS outperformed state-of-the-art DA methods like Cutout, Random Erasing, MixUp, and CutMix.
• Achieved high accuracy (97.56%) and AUROC (98.61%) on the IQ-OTH/NCCD dataset.
• Demonstrated superior performance (97.78% accuracy, 99.46% AUROC) on the chest CT scan images dataset.

Conclusions:

• The RPS technique significantly enhances CNN and transformer models for lung cancer diagnosis.
• RPS offers a promising approach for more accurate automated detection of lung cancer from CT scans.
• This advancement highlights the potential of AI in early lung cancer detection.