Lung Cancer Diagnosis from Computed Tomography Images Using Deep Learning Algorithms with Random Pixel Swap Data Augmentation

Back to abstracts
Abstracter Image
Click to view

ABSTRACT

  • 1Abe AA,
  • 2Nyathi M ,
  • 3Okunade A.A
  • 1Department of Medical Physics, Sefako Makgatho Health Science University, South Africa
  • 2Department of Medical Physics, Sefako Makgatho Health Science University, South Africa
  • 3Department of Physics and Engineering Physics, Obafemi Awolowo University, Nigeria

Background

Deep learning (DL) shows promise for automated lung cancer diagnosis, but limited clinical data restricts performance. While data augmentation (DA) helps, existing methods struggle with chest computed tomography (CT) scans across diverse DL architectures. This study proposes Random Pixel Swap (RPS), a novel DA technique to enhance diagnostic accuracy in both convolutional neural networks and Transformers for lung cancer detection from CT scans.

Method

RPS generates augmented data by randomly swapping pixels within CT scans. We evaluated it on ResNet, MobileNet, Vision Transformer, and Swin Transformer using two public CT datasets, measuring accuracy and area under the receiver characteristic curve (AUROC). Statistical significance was assessed via paired t-tests.

Result

The RPS outperformed state-of-the-art DA methods (Cutout, Random Erasing, Mixup, CutMix; p < 0.05), achieving 97.56% accuracy and 98.61% AUROC on IQ-OTH/NCCD and 97.78% accuracy (99.46% AUROC) on chest CT scans. While traditional augmentations (flipping, rotation) remained effective, RPS complemented them, surpassing prior studies and demonstrating artificial intelligence’s potential for early lung cancer detection.

Conclusion

RPS enhances CNN and Transformer models, enabling more accurate automated lung cancer detection in CT scans

Keywords: lung cancer diagnosis; deep learning; data augmentation; convolutional neural network; transformer; random pixel swap
Download Abstract
Author Image

PRESENTING AUTHOR

Mr. Ayomide Abe,

Student, Sefako Makgatho health Sciences University

Contact the author
Back to abstracts