Deep Convolution Neural Network for Laryngeal Cancer Classification on Contact Endoscopy-Narrow Band Imaging.

Nazila Esmaeili,Nikolaos Davaris,Michael Friebe,Alfredo Illanes,Elmer Jeto Gomes Ataide,Nassir Navab,Christoph Arens,Axel Boese,Esam Sharaf

doi:10.3390/s21238157

Abstract

(1) Background: Contact Endoscopy (CE) and Narrow Band Imaging (NBI) are optical imaging modalities that can provide enhanced and magnified visualization of the superficial vascular networks in the laryngeal mucosa. The similarity of vascular structures between benign and malignant lesions causes a challenge in the visual assessment of CE-NBI images. The main objective of this study is to use Deep Convolutional Neural Networks (DCNN) for the automatic classification of CE-NBI images into benign and malignant groups with minimal human intervention. (2) Methods: A pretrained Res-Net50 model combined with the cut-off-layer technique was selected as the DCNN architecture. A dataset of 8181 CE-NBI images was used during the fine-tuning process in three experiments where several models were generated and validated. The accuracy, sensitivity, and specificity were calculated as the performance metrics in each validation and testing scenario. (3) Results: Out of a total of 72 trained and tested models in all experiments, Model 5 showed high performance. This model is considerably smaller than the full ResNet50 architecture and achieved the testing accuracy of on the unseen data during the last experiment. (4) Conclusion: The proposed fine-tuned ResNet50 model showed a high performance to classify CE-NBI images into the benign and malignant groups and has the potential to be part of an assisted system for automatic laryngeal cancer detection.

Highlights

Laryngeal cancer is one of the most common malignancies in the head and neck area, with a growing incidence rate every year [1]
In order to account for the issue of overfitting of ResNet50, we proposed setting the cut-off-layer to discard part of the network
When compared to the metrics produced by Model 6 and Model 7, these scores were higher in both the validation and testing phases

Summary

Introduction

Laryngeal cancer is one of the most common malignancies in the head and neck area, with a growing incidence rate every year [1]. The endoscopic imaging modalities have become the standard procedure for screening and early diagnosis of laryngeal cancerous and precancerous lesions in clinical settings. These methods are widely applicable before performing a surgical biopsy for histological tissue examination in the context of the so-called optical biopsy [5,6]. The use of CE-NBI for diagnosis highly relies on the experience of the otolaryngologists and requires several years of training This can result in a subjective decision process followed by an overtreatment or undertreatment planning [7,9,10]

Objectives

Methods

Results

Discussion

Conclusion