DESMPA-Deep CNN: Double Exponential Smoothing Marine Predator Algorithm-Based Deep CNN for Liver Tumor Segmentation and Classification

Nowadays, liver cancers are well-known cancers attacking the human body. The larger segment of liver carcinomas is highly affected by alcohol-associated hepatitis as well as cirrhosis circumstances. In order to increase the survival rate of humans, premature and early diagnosis of liver cancer is very vital. Currently, discriminating the tumor and liver segments from medical imaging with the help of entirely automated computer-aided software is a highly difficult task, because the liver tumor may differ from one human being to another human being. Thus, this paper endeavors to present a new liver tumor segmentation and classification approach by incorporating an optimization-enabled deep learning model. Here, the adopted model comprises four stages, pre-processing, liver segmentation, liver tumor detection, and classification. To begin with, the collected images are given to the pre-processing stage using the African vulture's optimization-based adaptive histogram equalization (AVOAHE). The integration of the African vulture's optimization (AVO) concept in adaptive histogram equalization is named AVOAHE. After the image pre-processing phase, the liver region is segmented by utilizing kernel-based fuzzy C-means clustering. Then, the detection of liver tumor is performed using the Swin-Unet. Finally, tumor detection is done by utilizing the deep convolutional neural network (Deep CNN). Moreover, a hybrid optimization approach, called double exponential smoothing (DES)-marine predator algorithm (MPA), is developed by combining DES and MPA, which is used to train Deep CNN and serves as an enhancement. Finally, the experimental analysis has shown that the adopted technique has achieved better results in terms of accuracy, specificity, sensitivity, and precision with values of 0.957, 0.969, 0.941 and 0.915, respectively.