A quantitative comparison of nature inspired feature selection algorithms for diagnosis of liver HCC

Hepatocellular carcinoma (HCC) is the most prevalent form of liver cancer and the fourth leading cause of cancer-related deaths worldwide. The high mortality rate is largely attributed to limitations in effective screening and early diagnosis. This study presents a radiomics-based classification framework that diverges from conventional approaches by utilizing the entire liver Region Of Interest (ROI) rather than focusing solely on tumor-localized areas. This approach enables us to examine whether HCC has a global effect on the liver structure and alters its overall morphology, thereby allowing it to be distinguished from a healthy liver. Radiomic features are extracted from the liver ROI using both raw MRI data and the seven enhancement filters provided. These filters are employed to capture diverse texture patterns and enhance subtle image characteristics that may not be apparent in raw MRI data. To improve classification performance and eliminate irrelevant features, binary versions of four natureinspired optimization algorithms—African Vulture Optimization Algorithm (AVOA), Bat Algorithm (BA), Harris Hawks Optimization (HHO), and Sparrow Search Algorithm (SSA)— are applied for feature selection. We demonstrate that a publicly available dataset, commonly used for segmentation tasks, can also be effectively utilized for classification problems. This study demonstrates that high classification performance can be achieved even without tumor masks and the presence of HCC can be accurately detected using features extracted from the whole liver.