Lung Cancer Segmentation Using an Enhanced TransUNet++ Architecture

Read full text Summarise with VIRAT
Authors: C V Anoop, Karthik G.M., Dr Savitha S., Anil Kumar Muthevi, Agilesh Saravanan R, Ramu B.
Cited by: 0
References: 25

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Lung cancer is a life-threatening disease in which accurate staging of malignant nodules using computed tomography (CT) scans is critical for reducing mortality. Most existing approaches rely solely on deep learning models. This work proposes an accurate and computationally efficient hybrid deep learning framework for lung cancer analysis, integrating advanced preprocessing, feature extraction, and hybrid network architectures. The pipeline begins with preprocessing steps including resizing, normalization, edge detection, and median filtering to enhance image quality. Texture features are extracted using local binary patterns (LBP), while principal component analysis (PCA) is applied for dimensionality reduction. The optimized features are classified using an EfficientNet-B0 model. For precise segmentation, EfficientNet-B0 is embedded within a Transformer-based UNET++ (TransUNET++), enabling effective modeling of both local details and global contextual dependencies. Evaluated on a benchmark CT dataset, the proposed method achieved 98.58% accuracy, 98.47% sensitivity, 99.23% specificity, and 98.42% precision for classification, along with strong segmentation performance (99.53% Dice similarity coefficient, 98.56% Intersection over Union, 99.73% Hausdorff distance, 98.86% volumetric overlap error), demonstrating high spatial agreement with ground-truth masks.

Fields of Study & Subjects

AI in cancer detection Medical Image Segmentation Techniques Radiomics and Machine Learning in Medical Imaging

Publication details

Journal / source
Journal of Biomedical Photonics & Engineering
Volume
12
Issue
1
Pages
010305–010305
Publication date
23 March 2026
Language
EN