3D Texture Feature-Based Lymph Node Automated Detection in Head and Neck Cancer Analysis

Lymph node extracapsular extension (ECE) is a decisive indication for treatment planning of patients with head and neck squamous cell carcinoma (HNSCC). Lymph node region identification and segmentation is crucial for computer-aided ECE diagnosis. In this research, we propose a systematic machine learning approach to detect lymph node region from computed tomography (CT) scans based on 3D texture features. The process includes three steps: 1) region of interests (ROI) segmentation, where the potential lymph node region is segmented based on morphological operation; 2) 3D Haralick texture features are collected for lymph node and non-lymph node samples, respectively; and 3) three machine learning models with feature extraction and selection approaches are employed for lymph node classification. Based on 5-fold cross-validation, the experimental results have demonstrated that gradient boosting model with feature agglomeration and low variance threshold achieves the test accuracy of 94.48%. To check the explainability of the models, feature analysis is also conducted. The outcome of this research is expected to promote the implementation of artificial intelligence for lymph node detection as well as head and neck cancer diagnosis in the radiology computer vision field.