DefinePK

Abstract

In this study, a modified U-Net architecture was used to segment the laryngeal nodule completely automatically while being closely monitored. Diagnosing laryngeal cancer is difficult since the larynx is complicated and the illness only slightly alters it. Deep learning algorithms have shown promise in medical image processing, including the diagnosis of cancer. U-Net is a well-liked deep learning architecture for picture segmentation. The outcomes are compared to those obtained using other contemporary methods and the original U-Net that was published in order to arrive at the original conclusions.
In order to investigate how this affects the segmentation of Laryngeal nodules overall, this work additionally substitutes deconvolution layers for the up-sampling layers in both networks. During training, data augmentation was used right away. Initially, distortion and rotation combinations are used because of the poor quality of the source photos. This resulted in a limited data enrichment technique for combinations. Using the same parameter settings, the network is trained by substituting deconvolutional layers for all of the upsampling layers. There is greater dice overlap when comparing the suggested method for Laryngeal nodule segmentation to the state-of-the-art.