A Hybrid Inception Cross Covariance Image Transformer Framework for Robust Lung Nodule Classification from CT Images

Read full text Summarise with VIRAT
Authors: Sree Sankar. J, Manimegalai. P, P K Jayalakshmi
Cited by: 0
References: 26

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 nodules are small abnormal growths within lung tissue that indicate early-stage lung disease and require accurate differentiation from healthy regions for reliable clinical assessment.Accurate lung nodule classification from computed tomography (CT) images is crucial, as subtle morphological variations and low contrast complicate visual interpretation and reduce diagnostic consistency.Traditional diagnostic approaches rely on handcrafted feature extraction and radiologist driven analysis, which has constraints from interobserver variability, sensitivity to imaging artifacts and limited in large scale screening settings.This study offers a hybrid deep learning (DL) architecture that incorporates Inception V3 based multi scale convolutional feature extraction with a Cross Covariance Image Transformer (XCiT) for lung CT image classification.The Inception V3 extracts discriminative local spatial and textural features, while the XCiT module captures global channel wise contextual dependencies to enhance feature representation.The framework was evaluated on a balanced lung CT image dataset comprising 20,432 images using stratified training, validation and testing splits.Experimental results demonstrate high classification performance, attaining 99.27% accuracy, 99.21% precision, 99.31% recall and 99.26% F1 score.The framework exhibits stable convergence, minimal generalization limit and reliable discrimination between healthy and lung nodule cases, outperforming recent DL and transformer-based approaches.This study paves the way for an accurate, scalable and automated lung nodule classification approach that support clinical decision making and advance diagnostic reliability in computer aided lung imaging applications.

Fields of Study & Subjects

AI in cancer detection Radiomics and Machine Learning in Medical Imaging Lung Cancer Diagnosis and Treatment

Publication details

Journal / source
International journal of intelligent engineering and systems
Volume
19
Issue
4
Pages
566–585
Publication date
23 March 2026
Language
EN