Deep Feature Extraction with SVM Classification: A Robust Framework for Lung Cancer Detection in Histological Images

DOI

10.2991/978-94-6239-713-2_2How to use a DOI?

Keywords

Lung cancer; Histopathology; VGG16; Support Vector Machine; Dandelion Optimizer (DO); Deep learning; Stain normalization; Multimodal fusion; Digital pathology

Abstract

Lung cancer is the largest cause of cancer-related death globally, necessitating the urgent development of reliable and rapid diagnostic tools to aid clinical decision-making. Although deep learning models like DenseNet121 and ResNet-50 perform well in image-based classification tasks, their high tuning complexity, overfitting risk, and reliance on large-scale datasets restrict their usefulness in tiny and high-dimensional biomedical situations. Existing VGG16- and SVM-based studies have also shown promise, but most have been constrained by small sample numbers, insufficient hyperparameter optimization, and a dependence on single-modality data, pre-venting them from properly capturing the biological heterogeneity of lung cancer. To overcome these constraints, this paper offers a new Dandelion Optimizer (DO)-based hybrid VGG16-SVM architecture for multimodal lung cancer classification. The proposed system uses pretrained VGG16 for deep feature extraction because its uniform neural architecture retains fine-grained histopathological textures and allows for reliable multi-scale representation learning. SVM is utilized as a classifier to successfully handle high-dimensional feature spaces and increase generalization under limited-data situations, whereas DO is used to optimize critical SVM hyperparameters for better decision-boundary creation. Images from the LungHist700 dataset were preprocessed using Macenko stain normalization, background eradication, augmentation, and patch-level ex-traction at 224 × 224 resolution. The retrieved 512-dimensional deep features were combined with transcriptomic, proteomic, and clinical data to create a structured multimodal picture. This study’s originality resides in the integration of multimodal biological data with a decoupled VGG16-SVM architecture and DO-based optimization, which allows for better modelling of nonlinear interactions and cancer heterogeneity while improving robustness, interpretability, and generalization. Experimental findings show that the proposed framework outperforms conventional and unimodal baselines, with 96.94% accuracy, 97.11% precision, 98.95% recall, a 98.02% F1-score, and an AUC of 0.99. These results validate the proposed model as a scalable, interpretable, and high-performance framework for lung cancer classification, laying the groundwork for future clinically applicable AI-based diagnostic tools.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - B. Jayaprakash
PY  - 2026
DA  - 2026/06/25
TI  - Deep Feature Extraction with SVM Classification: A Robust Framework for Lung Cancer Detection in Histological Images
BT  - Proceedings of the International Conference on Advances in Computing Technology and Artificial Intelligence (COMPUTATIA 2026)
PB  - Atlantis Press
SP  - 4
EP  - 39
SN  - 2589-4919
UR  - https://doi.org/10.2991/978-94-6239-713-2_2
DO  - 10.2991/978-94-6239-713-2_2
ID  - Jayaprakash2026
ER  -