A Systematic Survey of Deep Learning Techniques for Liver Cancer Detection in Medical Imaging

Abstract

Introduction: Deep learning has become a dominant paradigm for analysing medical images in liver cancer detection, segmentation, and prognosis estimation. However, the rapid growth of research across diverse imaging modalities, tasks, and model architectures has resulted in fragmented findings, inconsistent evaluation practices, and limited comparability across studies. These challenges hinder clear understanding of methodological progress and limit the translation of research outcomes into clinical practice. Consequently, there is a strong need for a structured and systematic survey that consolidates existing work and provides a unified analytical perspective.

Objectives: The objective of this survey is to comprehensively analyse and synthesise state-of-the-art deep learning methods applied to liver cancer imaging. The study aims to organise prior research using well-defined taxonomy axes, identify dominant methodological and architectural trends, compare different learning paradigms, and highlight unresolved challenges related to robustness, generalisation, reproducibility, and clinical adoption.

Conclusions: This survey concludes that deep learning has substantially advanced liver cancer imaging, particularly through the adoption of hybrid, attention-based, and multimodal frameworks. Nevertheless, research progress remains fragmented due to dataset dependency, architectural complexity, and inconsistent evaluation practices. By consolidating existing studies into a unified taxonomy and critical synthesis, this work provides a reference framework and outlines key directions for future research toward robust, scalable, and clinically deployable liver cancer imaging systems.