Explainable Convolutional Neural Networks for Assessing Head and Neck Cancer Histopathology

Background Deep Learning algorithms show remarkable performance in the analysis of histopathological slides. We applied Convolutional Neural Networks and Explainable AI techniques to classify head and neck cancer.

Methods We manually annotated 101 histological slides of locally advanced head and neck squamous cell carcinoma. We trained neural networks to classify tumor and non-tumor tissue and to semantically segment four classes – tumor, non-tumor, non-specified tissue, and background. We studied features contributing to the networks’ decisions using Explainable AI methods, namely Grad-CAM and HR-CAM.

Results The classification network achieved an accuracy of 89.9% on previously unseen data. Our segmentation network achieved a class-averaged Intersection over Union score of 0.690, and 0.782 for tumor tissue in particular. Explainable AI methods suggested that nuclear features highly contributed to tumor predictions, which agree with features used by pathologists.

Conclusions Our work shows that neural networks can predict head and neck cancer with high accuracy. They show great potential for assisting pathologists in the assessment of head and neck cancer histopathology, especially if their predictions are explained visually.

This work was funded in part by the Federal Ministry of Education and Research (BMBF) to AG and ME (01KD2211B) and to AMK (01KD2211A).

Publication History

Article published online:
12 May 2023