A Robust Deep Learning Algorithm for Lung Cancer Detection from Computed Tomography Images

Back to abstracts
Abstracter Image
Click to view

ABSTRACT

  • 1Nyathi M,
  • 2Abe AA ,
  • 3Okunade AA ,
  • 4Pilloy W ,
  • 5Kgole B ,
  • 6Nyakale N
  • 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.
  • 4Department of Nuclear Medicine, Dr George Mukhari Academic Hospital
  • 5Department of Internal Medicine, Doctor George Mukhari Academic Hospital
  • 6Department of Nuclear Medicine, Dr George Mukhari Academic Hospital

Background

Early detection of lung cancer from chest computed tomography (CT) scans improves cure rates, but subtle early-stage patterns can challenge radiologists. While deep learning (DL) models aid diagnosis, they typically require large datasets.

Method

This study proposes an ensemble convolutional neural network trained on a relatively small dataset (IQ-OTH/NCCD) to automate lung cancer detection. The potential of lung region of interest (ROI) segmentation and data augmentation was assessed for the ensemble model.

Result

The results showed that the ensemble model surpassed each individual CNN. The data augmentation and ROI segmentation further enhanced performance achieving 98.17% accuracy, 98.21% sensitivity and 98.13% specificity in distinguishing cancerous vs. non-cancerous scans and 95.43% accuracy, 93.40% sensitivity and 97.09% specificity in classifying normal, benign, or malignant nodules.

Conclusion

Using ensemble techniques, segmentation of the lung ROI and data augmentation enhances the diagnostic performance of DL models when data is limited. This approach demonstrates DL’s potential to enhance early diagnosis and support radiologists.

Keywords: Lung cancer, early diagnosis, computed tomography, deep learning, ensemble model.
Download Abstract
Author Image

PRESENTING AUTHOR

Prof. Mpumelelo Nyathi, PhD Medical Physics

Associate Professor, Sefako Makgatho health Sciences University

Prof Mpumelelo Nyathi holds a PhD in Medical Physics. He coordinates the Bachelor of Medicine and Bachelor of Surgery Extended Degree Program (MBChB-ECP) at Sefako Makgatho Health Sciences University. Prof Nyathi is also extensively involved in research and supervision of postgraduate students up to PhD level in the Department of Medical Physics at SMU. He serves as a peer reviewer for national and international journals. Prof Nyathi has authored over 30 publications in peer-reviewed journals and has presented in national and international Medical Physics conferences. His research interests include interactions of ionizing radiation with biological matter, medical imaging, radiation protection and application of artificial intelligence in medical physics.
Contact the author
Back to abstracts