Purpose: In medical image-based computer-aided diagnosis, various artificial intelligence (AI)-based models have been proposed and demonstrated high performance. However, there is still a gap between the human’s understanding and how these models learned, which influence how we understand and trust these AI algorithms. Therefore, the interpretability of AI model is essential. Our objective is to automatically perform Extracapsular Extension (ECE) detection with an interpretable deep learning technique. ECE is a decisive indication for treatment planning of patients with head and neck squamous cell carcinoma (HNSCC). Current detection practice often requires the visual identification of professionals and ultimately pathologic confirmation. We investigate AI techniques to assist with human identification.
Methods: In this research, we propose a deep learning method to detect ECE from 3D computed tomography (CT) scans. The proposed model has multi-input that captures information with different scales. The deep convolutional neural network (DCNN) structure has been implemented. To check the explainability of the model, 3D possibility map is utilized to highlight the possible ECE region. Then, the results can be compared with the locations of lymph node region.
Results: The experimental results have demonstrated that our model can identify ECE and non-ECE patients. Specifically, we have achieved the ECE detection with 96.92% accuracy, and 98.84% AUC. In addition, ECE probability maps are generated for patients, and ECE high potential regions are shown, which is able to enhance the interpretability of the model.
Conclusion: The research demonstrates the capability to use artificial intelligence for ECE detection. By introducing the possibility map, we can identify where ECE has possibly occurred. Therefore, this will contribute to the explainability of the proposed model. The outcome of this study is expected to promote the implementation of interpretable artificial intelligence-assist ECE detection.
IM/TH- Image Analysis (Single Modality or Multi-Modality): Computer-aided decision support systems (detection, diagnosis, risk prediction, staging, treatment response assessment/monitoring, prognosis prediction)