A Review of Recent Efficient Deep Learning Methods for Liver Segmentation

Abstract

Introduction: The accurate and timely diagnosis of non-alcoholic fatty liver disease (NAFLD) and precise segmentation of tumors and liver are significant and critical for successful management of patients and improved medicinal outcomes. This review consolidates current developments in methodologies and algorithms used in medical imaging, particularly detection and segmentation of such intricate liver diseases. It explores the wide range of diagnostic techniques that extend from old school machine learning methodologies, such as classification tree-based methods of diagnosing NAFLD [1], to highly advanced deep learning models crafted specifically for analysis of ultrasound as well as computed tomography (CT) images. Specifically, we examine the application of deep learning models for ultrasound image classification in NAFLD and the development of advanced segmentation frameworks utilizing self-ONN-based decoders, U-Net architectures [6], and graph convolutional networks [8] for precise liver and tumor delineation. The availability of large-scale and carefully labeled medical image collections, such as the dataset created by Alshagathrh and colleagues [4], has been crucial for teaching and testing these advanced computer models. This has led to notable progress in how accurately we can diagnose liver conditions and precisely locate problems within the liver [5, 7]. This review brings together these advancements, highlighting how these computer-based methods have the potential to significantly improve the way we understand and manage liver diseases, ultimately leading to more efficient and dependable healthcare practices.

Objectives: The key objective is to bring together the advancements in recent efficient deep learning models, highlighting how these computer-based methods have the potential to significantly improve the way we understand and manage liver diseases, ultimately leading to more efficient and dependable healthcare practices.

Conclusions: In this review, the changing picture of liver disease segmentation and diagnosis, specifically the NAFLD, and tumor liver analysis via recent machine learning and deep learning strategies have been reviewed. Earlier contributions to this work were dependent upon rule-based statistical and classic procedures like trees in classification in the identification of factors that contribute to NAFLD [1]. With increasing computational power and sizes of medical image datasets, the paradigm has undoubtedly shifted towards encoder-decoder architecture and convolutional neural networks (CNNs), including UNet and its several variants [6][7][8].