Suspected region segmentation and multi-classification for gastrointestinal disease with deep learning model using endoscopy image
TL;DR
A new technique called Transit Crayfish Optimization_Deep Residual Network (TCFO_DRN), which is trained by Transit Crayfish Optimization (TCFO) and obtained by the integration of the Crayfish Optimization Algorithm (COA) and Transit Search (TS) approach.
Abstract
Gastrointestinal (GI) syndromes have a significant impact on life expectancy and a high mortality rate. It is difficult to analyze vast amounts of medical image data, which raises the possibility of incorrect medical assessments. Therefore, this paper developed a new technique called Transit Crayfish Optimization_Deep Residual Network (TCFO_DRN). At first, an endoscopy image of the GI is considered as input, and the noise present in the image is eradicated by the Medav Filter. Then, the segmentation of the suspected region is carried out by employing a hybrid segmentation approach comprising Fully Convolutional Networks (FCN) and the Recurrent Prototypical Network (RP-Net). The outputs from both segmentation approaches are fused using the Pearson correlation coefficient. Following this, feature extraction is performed, and then GI classification is executed by utilizing a Deep Residual Network (DRN), which is trained by Transit Crayfish Optimization (TCFO). Here, TCFO is obtained by the integration of the Crayfish Optimization Algorithm (COA) and Transit Search (TS) approach. Furthermore, the efficiency of TCFO_DRN is assessed using the Kvasir dataset and HyperKvasir dataset, where TCFO_DRN computed a maximum accuracy of 0.950, True Positive Rate (TPR) of 0.962, and True Negative Rate (TNR) of 0.933 using the HyperKvasir dataset.