Back to all papers

A novel ensemble transfer learning approach for lung cancer classification using advance VGGNet16 with wavelet transform equalization & CL-PSO.

December 9, 2025pubmed logopapers

Authors

Das MN,Panda N,Rautray R,Tripathy J,Moreira F

Affiliations (5)

  • Dept. of Computer Science and Engineering, ITER(FET), Siksha 'O' Anusandha (Deemed to be) University, Bhubaneswar, India. Electronic address: [email protected].
  • Dept. of Computer Science and Engineering, ITER(FET), Siksha 'O' Anusandha (Deemed to be) University, Bhubaneswar, India. Electronic address: [email protected].
  • Dept. of Computer Science and Engineering, ITER(FET), Siksha 'O' Anusandha (Deemed to be) University, Bhubaneswar, India. Electronic address: [email protected].
  • Dept. of CSE-AIML & IoT, Vallurupalli Nageswara Rao Vignana Jyothi Institute of Engineering and Technology, Hyderabad, India. Electronic address: [email protected].
  • REMIT, IJP, Universidade Portucaelense, Porto & IEETA, Universidade de Aveiro, Aveiro, Portugal. Electronic address: [email protected].

Abstract

Lung cancer poses severe burden to the world and well-yield and requiring high-yield and easily deployable diagnostic strategies. This study proposes an enhanced deep learning approach for early lung cancer diagnosis using a fine-tuned VGG-16 model optimized with Comprehensive Learning Particle Swarm Optimization (CL-PSO). To mitigate data imbalance and enhance feature visibility in CT scans, the framework introduces the Wavelet Transform Equalization in the preprocessing and utilizes class-weighted training to improve detection sensitivity, especially for the underrepresented benign cases. The model scored almost perfect classifications of the IQ-OTH/NCCD dataset with an accuracy of 99.99 %, precision and recall of 99.98 %, F1-score of 99.99 % and the AUC-ROC of 1.00. Grad-CAM visualizations further enhanced the model's interpretability and confirming that its predictions corresponded with radiological decision points. Apart from that, the model responded robustly to noise, occlusion, illumination, and below 50 ms per image. This results making model an ideal for real-time integration into imager based hospital PACS and edge based healthcare systems.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 7,100+ peers who rely on RadAI Slice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.