Development of a Hybrid Methodology of Deep Learning and Machine Learning for Lung Nodule Detection in Medical Computed Tomography Images

Authors

Zaed S. Mahdi, University of Technology - Iraq, Information Technology Center Al-Sina'a St., Al-Wehda District, 10066 Baghdad, IraqFollow
Rana M. Zaki, University of Technology - Iraq, Department of Computer Science Al-Sina'a St., Al-Wehda District, 10066 Baghdad, IraqFollow
Alaa Kadhim Farhan, University of Technology - Iraq, Department of Computer Science Al-Sina'a St., Al-Wehda District, 10066 Baghdad, IraqFollow
Negar Majma, Naghshejahan Higher Education Institute, Computer Science Department Isfahan, IranFollow

Abstract

Deep learning and machine learning play an important role in the medical field, helping doctors make accurate, fast and effective diagnosis. Despite the progress achieved in the use of modern technologies in detecting cancerous nodes, current studies still suffer from some challenges and limitations that must be addressed to obtain high efficiency in identifying cancerous nodes. These challenges include using image pre-processing, combining deep learning and machine learning techniques, and constantly adapting to clinical changes, in order to address this. A hybrid methodology has been proposed for detecting cancerous nodules in the lung in medical Computed Tomography (CT) images. It includes three stages, the first of which is pre-processing the images used, identifying nodes, balancing samples, and extracting features using Local Binary Pattern (LBP) and Histogram of Oriented Gradients (HOG) from the images, and the second is based on building a deep learning model consisting of Convolutional Neural Network (CNN) and Long-Short-Term Memory (LSTM) to extract the trained features and combine them with the features extracted by LBP and HOG to be input for the next stage. In the final stage, the improved eXtreme Gradient Boosting (XGBoost) machine learning classifier is built using the Particle Swarm Optimization (PSO) algorithm. The highest accuracy results of 99.3%, and ROC-AUC were obtained 99.8%. The proposed methodology has proven its efficiency in detecting cancerous nodes accurately using CT images.

Recommended Citation

Mahdi, Zaed S.; Zaki, Rana M.; Farhan, Alaa Kadhim; and Majma, Negar (2024) "Development of a Hybrid Methodology of Deep Learning and Machine Learning for Lung Nodule Detection in Medical Computed Tomography Images," Journal of Soft Computing and Computer Applications: Vol. 1: Iss. 2, Article 1011.
DOI: https://doi.org/10.70403/3008-1084.1011