A Novel PHOG Feature for Lung Nodule Characterization

Lung cancer remains one of the leading causes of cancer-related mortality worldwide, primarily due to its asymptomatic nature in early stages and the resulting late diagnoses. Early and accurate characterization of pulmonary nodules plays a critical role in improving patient outcomes by enabling timely intervention. However, distinguishing benign from malignant nodules remains a significant clinical challenge, often leading to unnecessary biopsies and delayed treatments. To address this issue, a novel machine learning (ML) framework has been developed by introducing the Pyramid Histogram of Oriented Gradients (PHOG) feature for nodule classification. The proposed model effectively extracts and learns both spatial and shape-related information of lung tumors through multi-level gradient orientation histograms, capturing the intricate structural details of nodules using Support Vector Machine (SVM) classifier. The proposed framework has been validated on publicly available LIDC-IDRI dataset using k-fold cross validation technique and achieved average sensitivity, specificity, and accuracy of 96.21%, 94.89%, and 95.46% respectively for nodule characterization. The introduced multi-scale PHOG feature based framework has been found to outperform other state-of-the-art models, indicating its significant potential for reliable lung nodule characterization in clinical application.