Optimal edge optimization and deep learning model for liver tumour nuclei segmentation and classification

This paper develops an Optimal Edge Segmentation and Deep Learning Model (OESDLM) for achieving optimal nuclei segmentation and classification of liver tumour. The input histopathology is first collected from internet resources. Subsequently, pre-processing techniques including colour normalization, histogram equalization, and median filtering are taken into consideration to eliminate noise from the input images. Next, a novel unsupervised edge detection method remains applied to segment the nuclei areas in liver cancer histopathology images, thereby achieving the nuclei segmentation from the pre-processing. Rather than computing gradients, the idea of calculating local standard deviation is taken into consideration in this edge detection technique. Finally, an Adaptive Convolutional Neural Network (ACNN) is utilized to classify liver tumors. The Golf Optimization Algorithm (GOA) is taken into consideration when choosing the hyperparameters in the ACNN. Using performance metrics such as the Dice coefficient, accuracy, RMSE, accuracy, F_Score, ROC, Recall, Precision, and Accuracy, the suggested methodology is put into practice in MATLAB.