A dense kernel point convolutional neural network for chronic liver disease classification with hybrid chaotic slime mould and giant trevally optimizer

Abstract

Chronic liver disease affects liver tissues and can lead to liver failure. Early diagnosis is crucial for providing better treatment and reducing the mortality rate. Traditional methods like biopsy and manual analysis of liver computed tomography scan images are commonly used for diagnosis. However, biopsy is invasive and can cause pain and other complications, while manual analysis by a radiologist is time-consuming and requires expert knowledge. This manuscript presents the Dense Kernel Point Convolutional Neural Network for Chronic Liver Disease Classification with Hybrid Chaotic Slime Mould and Giant Trevally Optimizer (DKPCNN-CLDC-HybCSM-GTO) to improve diagnostic performance. For image segmentation, Fuzzy C-Ordered Means Clustering (IFCMC) method is applied, and for feature extraction, the Intuitionistic Invariant Wavelet Scattering Transform (IWST) is employed. The DKPCNN-CLDC-HybCSM-GTO method is compared with existing methods, including Preoperative Classification of Primary and Metastatic Chronic Liver Disease via Machine Learning-Based Ultrasound Radiomics (PC-PM-LC-MLUR), the Liver Disease Classification from Ultrasound Using Multi-Scale CNN (LDC-US-MS), and the Convolutional Neural Network for Classifying Primary Chronic Liver Disease Based on Triple-Phase CT and Tumor Marker Information (CNN-CPLC-TPCT-TMI). The evaluation metrics include accuracy, precision, sensitivity, specificity, and F1 score. The DKPCNN-CLDC-HybCSM-GTO approach shows significant improvements over existing methods: 22.36%, 25.42%, and 18.27% higher accuracy; 22.36%, 15.42%, and 18.27% higher sensitivity; and 21.36%, 16.42%, and 19.27% higher specificity, respectively. These results demonstrate the method being proposed offers a more a valuable tool for early detection and better treatment outcomes in chronic liver disease.