Early Detection of Laryngeal Cancer using Multiple Instance Learning Based Neural Network

Laryngeal cancer is the most common type of head and neck cancer which affects the soft tissues of the larynx. Early stage detection of laryngeal cancer is crucial to avoid further medical complications and better patient care. The primary aim of this research is to provide computer-aided cancer diagnosis powered by deep learning mechanism. To achieve this we develop a novel Multiple Instance Learning (MIL) technique which classifies healthy and unhealthy/cancerous tissues. Further, we also incorporate the traditional Convolutional Neural Network (CNN) and transfer learning DenseNet121 model on our dataset for better comparison and evaluation of our research. The models are evaluated using standard metrics which are specific to biomedical domain. Our proposed MIL architecture produced outstanding results when compared to other models. It also outperformed other state-of-the-art MIL models used in solving medical domain problems. The results imply that the proposed technique is highly effective in detecting early signs of laryngeal cancer. The results prove that this proposed approach can assist medical professionals in early and accurate diagnostic of laryngeal cancer