Deep Learning-Based Detection of Pulmonary Involvement in Malignant Disease Using CT and 18F-FDG PET/CT Scans

Fallahpoor, Maryam

Deep Learning-Based Detection of Pulmonary Involvement in Malignant Disease Using CT and 18F-FDG PET/CT Scans

Fallahpoor, Maryam

Permalink

Publication Type:: Thesis
Issue Date:: 2025

Open Access

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is open access.

Adobe PDF

Download thesisAdobe PDF (4.99 MB)

View statistics

Full metadata record

Field	Value	Language
dc.contributor.author	Fallahpoor, Maryam
dc.date.accessioned	2025-09-16T03:36:53Z
dc.date.available	2025-09-16T03:36:53Z
dc.date.issued	2025
dc.identifier.uri	http://hdl.handle.net/10453/189894
dc.description	University of Technology Sydney. Faculty of Engineering and Information Technology.	en_US.UTF-8
dc.description.abstract	Pulmonary involvements, including pulmonary nodules, pose a risk of lung complications and lung cancer, the leading cause of cancer-related deaths globally. Diagnosis and categorization of pulmonary nodules is critical for lung cancer detection and staging other malignancies due to the high incidence of metastasis in the lungs from various cancers, which is typically performed using common medical imaging such as ¹⁸F FDG PET/CT imaging. Diagnosis of pulmonary involvement require significant experience, are labor-intensive, and prone to error. Artificial intelligence (AI), particularly deep learning, has shown promising results in medical applications. This study aims to develop a reliable 3D deep learning-based approach to assist physicians in accurately diagnosing pulmonary nodules from ¹⁸F FDG PET/CT imaging. We investigated lung nodules originating from malignancies other than lung cancer and excluded lung masses. Most studies focus only on lung nodules from lung cancer, with limited attention given to classifying nodules from other malignancies. The study has three main objectives: First, classification of pulmonary involvements, including COVID-19, using optimized deep learning models, generalizability assessment, and identification of the optimal pre-processing steps. Second objective is classification of pulmonary nodules from a subset of a ¹⁸F FDG PET/CT dataset using the state-of-the-art deep learning models to determine the best model and approach for multi-class classification of pulmonary nodules similar to real clinical scenarios. And the third objective is to develop an optimized deep learning-based approach for higher accuracy in pulmonary nodule detection and classification using a large ¹⁸F FDG PET/CT dataset. For the first objective, classification of lung involvement from COVID-19 was performed using four datasets of CT images and seven deep learning models in 5-fold cross-validation. To achieve the second objective, different deep learning models were tested on a subset of PET/CT dataset in a multi-class classification approach to refine models and methodologies through a trial-and-error approach. Finally, adjustments were made to the top-performing model's layers and pre-processing steps for optimization of lung nodule classification from 1304 ¹⁸F FDG PET/CT images. The results of this study indicated that the combining 80% of one dataset with at least 40% from another yields comparable results to using the full combination. We have achieved overall accuracy of 89% for lung nodule classification. The multi-class simulation, classifying benign, malignant, and suspicious cases, closely mimics real-world conditions, demonstrating the capability of the model to handle complex scenarios and its potential to assist medical professionals in making more accurate diagnoses.	en_US.UTF-8
dc.format	Thesis (PhD)
dc.language.iso	en_US	en_US.UTF-8
dc.relation	https://opus.lib.uts.edu.au/bitstream/10453/189894/1/thesis.pdf
dc.rights	info:eu-repo/semantics/openAccess
dc.rights	The author owns the copyright in this thesis including all reproduction and reuse rights for the work. The work may not be altered without the permission of the copyright owner. Attribution is essential when quoting or paraphrasing from this thesis.
dc.rights	© 2025 Maryam Fallahpoor
dc.rights	au.edu.uts.lib/cph
dc.title	Deep Learning-Based Detection of Pulmonary Involvement in Malignant Disease Using CT and 18F-FDG PET/CT Scans	en_US.UTF-8
dc.type	Thesis
utslib.copyright.status	open_access	*

Abstract:

Pulmonary involvements, including pulmonary nodules, pose a risk of lung complications and lung cancer, the leading cause of cancer-related deaths globally. Diagnosis and categorization of pulmonary nodules is critical for lung cancer detection and staging other malignancies due to the high incidence of metastasis in the lungs from various cancers, which is typically performed using common medical imaging such as ¹⁸F FDG PET/CT imaging. Diagnosis of pulmonary involvement require significant experience, are labor-intensive, and prone to error. Artificial intelligence (AI), particularly deep learning, has shown promising results in medical applications. This study aims to develop a reliable 3D deep learning-based approach to assist physicians in accurately diagnosing pulmonary nodules from ¹⁸F FDG PET/CT imaging. We investigated lung nodules originating from malignancies other than lung cancer and excluded lung masses. Most studies focus only on lung nodules from lung cancer, with limited attention given to classifying nodules from other malignancies. The study has three main objectives: First, classification of pulmonary involvements, including COVID-19, using optimized deep learning models, generalizability assessment, and identification of the optimal pre-processing steps. Second objective is classification of pulmonary nodules from a subset of a ¹⁸F FDG PET/CT dataset using the state-of-the-art deep learning models to determine the best model and approach for multi-class classification of pulmonary nodules similar to real clinical scenarios. And the third objective is to develop an optimized deep learning-based approach for higher accuracy in pulmonary nodule detection and classification using a large ¹⁸F FDG PET/CT dataset. For the first objective, classification of lung involvement from COVID-19 was performed using four datasets of CT images and seven deep learning models in 5-fold cross-validation. To achieve the second objective, different deep learning models were tested on a subset of PET/CT dataset in a multi-class classification approach to refine models and methodologies through a trial-and-error approach. Finally, adjustments were made to the top-performing model's layers and pre-processing steps for optimization of lung nodule classification from 1304 ¹⁸F FDG PET/CT images. The results of this study indicated that the combining 80% of one dataset with at least 40% from another yields comparable results to using the full combination. We have achieved overall accuracy of 89% for lung nodule classification. The multi-class simulation, classifying benign, malignant, and suspicious cases, closely mimics real-world conditions, demonstrating the capability of the model to handle complex scenarios and its potential to assist medical professionals in making more accurate diagnoses.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/189894