Title: An efficient deep learning approach for identifying interstitial lung diseases using HRCT images

Authors: Nidhin Raju; D. Peter Augustine

Addresses: Department of Computer Science, CHRIST (Deemed to be University), Bangalore, India ' Department of Computer Science, CHRIST (Deemed to be University), Bangalore, India

Abstract: Interstitial lung disease (ILD) encompasses over 200 fatal lung disorders affecting the interstitium, leading to significant mortality rates. We propose an AI-driven approach to diagnose and classify ILD from high-resolution computed tomography (HRCT) images. The research utilises a dataset of 3,045 HRCT images and employs a two-tier ensemble method that combines various machine learning (ML) models, convolutional neural networks (CNNs), and transfer learning. Initially, ML models achieve high accuracy, with the J48 model at 93.08% accuracy, mainly highlighting the importance of diagonal-wise standard deviation. Deep learning techniques are then applied, with three CNN models achieving test accuracies of 94.08%, 92.04%, and 93.72%. Transfer learning models also show promise, with InceptionV3 at 92.48% accuracy. Ensembling these models further boosts accuracy, with the ensemble of three CNN models reaching 97.42%. This research has the potential to advance ILD diagnosis, offering a robust computational framework that enhances accuracy and ultimately improves patient outcomes.

Keywords: interstitial lung disease; ILD; deep learning; DL; transfer learning; multi-label classification; high-resolution computed tomography; HRCT.

DOI: 10.1504/IJCSE.2024.138427

International Journal of Computational Science and Engineering, 2024 Vol.27 No.3, pp.286 - 301

Received: 19 Oct 2022
Accepted: 24 Jun 2023
Published online: 03 May 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article