Extraction of Laryngeal Cancer Informative Frames from Narrow Band Endoscopic Videos
Noha A. Sobhi, Sherin Moustafa Youssef, Marwa Ali Elshenawy · 2021
Laryngeal cancer is one of the most common types of throat cancer. One of the most powerful technologies is a Narrow-band imaging (NBI) endoscope, which helps in diagnosing the early stage of cancer and reducing the biopsy risks. However, reviewing an endoscopic video is a labor-intensive process, as it contains a large number of uninformative frames due to the illumination or reflection effect and the appearance of saliva. This paper aims at designing and implementing an enhanced automated model to select the informative Laryngoscope video frames to reduce the computational time of scanning all the frames. Also, the selection of the informative frame will help the specialist in the diagnosis process. The proposed model uses a set of quality assessment features including texture and color features. Texture features have been used to detect the sharpness of the image as it is an important measure of image clarity. Moreover, the color features will help in identifying the images with saliva or specular reflections. Then, the extracted features are fed to different classifiers such as Support Vector Machine (SVM) and different ensemble classifiers. The classifiers have been used to classify the video frame into one of four types (informative (I), blurred (B), saliva or specular reflections (S), and underexposed (U)). The experimental results show that the Random Forest (RF) classifier produced a very promising classification result, with an average classification recall equals to 95.8%. The proposed model obtained better classification recall by 2.2% compared to the existing state-of-the-art method.