A Gradient Mapping Guided Explainable Deep Neural Network for Extracapsular Extension Identification in 3D Head and Neck Cancer Computed Tomography Images

TL;DR

This paper introduces GMGENet, an explainable deep learning framework that automatically detects extracapsular extension in head and neck cancer CT scans without manual lymph node annotation, achieving high accuracy and aiding clinical decision-making.

Contribution

The study presents a novel gradient mapping guided explainable neural network that eliminates the need for manual lymph node annotation in ECE detection from CT images.

Findings

Achieved 90.2% accuracy and 91.1% AUC in ECE detection.

Guided focus on relevant regions improves interpretability.

Correlated ECE detection with histopathological findings.

Abstract

Diagnosis and treatment management for head and neck squamous cell carcinoma (HNSCC) is guided by routine diagnostic head and neck computed tomography (CT) scans to identify tumor and lymph node features. Extracapsular extension (ECE) is a strong predictor of patients' survival outcomes with HNSCC. It is essential to detect the occurrence of ECE as it changes staging and management for the patients. Current clinical ECE detection relies on visual identification and pathologic confirmation conducted by radiologists. Machine learning (ML)-based ECE diagnosis has shown high potential in the recent years. However, manual annotation of lymph node region is a required data preprocessing step in most of the current ML-based ECE diagnosis studies. In addition, this manual annotation process is time-consuming, labor-intensive, and error-prone. Therefore, in this paper, we propose a Gradient Mapping Guided Explainable Network (GMGENet) framework to perform ECE identification automatically without requiring annotated lymph node region information. The gradient-weighted class activation mapping (Grad-CAM) technique is proposed to guide the deep learning algorithm to focus on the regions that are highly related to ECE. Informative volumes of interest (VOIs) are extracted without labeled lymph node region information. In evaluation, the proposed method is well-trained and tested using cross validation, achieving test accuracy and AUC of 90.2% and 91.1%, respectively. The presence or absence of ECE has been analyzed and correlated with gold standard histopathological findings.