ItemPhD

Artificial Intelligence Based Computational Framework for Identification and Classification of Interstitial Lung Diseases Using HRCT Images

Title
Artificial Intelligence Based Computational Framework for Identification and Classification of Interstitial Lung Diseases Using HRCT Images
Creator
Nidhin, Raju
Contributor
Peter Augustin,D
Description
Interstitial Lung Diseases (ILDs) refer to a wide array of respiratory disorders characterised via infection and scarring of the lung's interstitial tissue. These conditions affect the space within the air sacs, compromising the lungs' ability to expand and contract properly. ILDs manifest with a range of symptoms, including persistent cough, shortness of breath, and fatigue. Diagnosis of ILDs often involves imaging methods, mainly High-Resolution Computed Tomography (HRCT), to assess lung abnormalities. ILDs can have lasting effects on respiratory function, leading to progressive fibrosis. The primary obstacle in identifying ILDs lies in the diverse array of symptoms they present, making it challenging to distinguish them from other pulmonary disorders. The HRCT is a commonly employed method in ILD diagnosis. These images provide a detailed depiction of lung tissue, revealing its size, shape, and any notable abnormalities or changes. Moreover, HRCT plays a crucial role in monitoring disease progression over time. Deep Learning (DL) excels in detecting patterns in intricate medical images that may pose challenges for traditional methods. Moreover, DL algorithms exhibit the ability to identify subtle changes in medical images indicative of pathology, and they can automate object detection tasks. The application of DL in medical contexts can enrich the precision and rapidity of diagnoses. In this research aimed at improving the accuracy of artificial intelligence AI-based ILD identification, we harnessed the benefits of deep learning, employing full-training, Transfer Learning (TL), and ensemble voting techniques. Our approach involved the construction of three Convolutional Neural Networks (CNNs) from scratch for ILD detection. Additionally, we customized models named InceptionV3, VGG16, MobileNetV2, VGG19, and ResNet50 for both full-training and TL strategies. This comprehensive methodology aimed to take benefits of DL architectures to enhance the precision of ILD identification in medical imaging. Both the first dataset consisting of HRCT images and the second dataset comprising Chest X-ray were employed in our study. However, during the initial training phase of the TL models, we utilized pre-trained ImageNet weights. To enhance performance, modifications were made to the classification layers of all five models for both TL and full-training processes. To further improve training outcomes, a soft-voting ensemble approach was employed. The ensemble, combining the predictions of all three newly developed CNN models (ILDNetV1, ILDNetV2 and ILDNetV3), and ILDNetV1 achieved the highest test accuracy at 98.14%. Additionally, we incorporated machine learning (ML) models, including Logistic Regression, BayesNet, RandomForest, Multilayer Perceptron (MLP), and J48, using statistical measurements derived from HRCT images. Our study introduces a novel AI-based system for predicting ILD categories. This system demonstrated superior performance on unseen data by leveraging the results from the newly constructed CNNs, transfer learning, and ML models. This comprehensive approach holds promise for advancing ILD category prediction, providing a more robust and accurate tool for medical diagnosis and decision- making.
Source
Author's Submission
Date
2024-01-01
Publisher
Christ(Deemed to be University)
Subject
Computer Science
Rights
Open Access
Relation
61000346
Format
PDF
Language
English
Type
PhD

Collection

PHD

Citation

Nidhin, Raju, “Artificial Intelligence Based Computational Framework for Identification and Classification of Interstitial Lung Diseases Using HRCT Images,” CHRIST (Deemed To Be University) Institutional Repository, accessed February 23, 2025, https://archives.christuniversity.in/items/show/12392.

Output Formats