Liver cancer classification using archery learning and enhanced bilinear convolutional neural network

Abstract

Liver Cancer (LC) remains as the primary reason of cancer based mortality. Early identification of LC improves the chances of successful treatment and patient recovery. Among different imaging modalities, Computed Tomography (CT) is generally exploited for detecting and diagnosing liver abnormalities. But, manual liver segmentation and associated tumors is manual process, takes more time and also prone to inconsistencies and subjective errors among radiologists. To address these challenges, Computer Aided Detection (CAD) have been presented for supporting clinicians. This work suggests an automated Deep Learning (DL) model for segmentation and classification of LC using segmentation and classification approaches. The input LC images pre-processed by the Gaussian filtering and the segmentation of the lesions are done by the Fuzzy C means based Archery Learning Algorithm (FALA). At last, the lesions are Multi-Head Attention based Bilinear Convolutional Neural Network (MHA-BCNN). The experimentation is demonstrated on two benchmark datasets and attained better accuracies of 99.5% (CT liver) and 99.2% (Primary Liver Cancer). This work delineates the ability of DL approach in ensuring robust and reliable tools for CAD of LC.