Search

Purpose/Objective(s) The extracapsular extension (ECE) is a strong predictor of patients' survival outcomes with head and neck squamous cell carcinoma (HNSCC). ECE occurs when metastatic tumor cells within the lymph node break through the nodal capsule into surrounding tissues. It is crucial to identify the occurrence of ECE as it changes staging and management for the patients. Current clinical ECE detection relying on radiologists' visual identification is extremely labor-intensive, time-consuming, and error-prone, and consequently, pathologic confirmation is required. Therefore, we aim to perform ECE identification automatically using a deep learning-based technique to analyze the presence or absence of ECE and correlate that with gold standard histopathological findings. Materials/Methods This research proposes a novel deep learning method to detect ECE from 3D computed tomography (CT) scans. The proposed network has multi-scale input that captures both local and global information. The two types of inputs are fitted into two pathways with deep convolutional neural networks (DCNNs), 3D CNN baseline and 3D DenseNet. A sliding-cube approach is applied to extract small paired 3D patches from patient data with different scales at data preparation. The network will classify the patient based on patch-level classification results. Different training scenarios are designed for the experimental test. Results Based on five-fold cross-validation, the experimental results have demonstrated that our model can identify ECE and non-ECE patients, specifically in the patch-level. We have achieved the ECE detection with 96.92% accuracy 98.84% AUC. Detailed results are shown in the Table 1. Conclusion The study demonstrates the ability to use a deep learning-based method for ECE direct detection. The proposed local-global network can help capture sufficient features and achieve high classification accuracy. The outcome of this research is expected to promote the implementation of artificial intelligence for ECE identification for head and neck cancer diagnosis in the radiology computer vision field.