Global trends in artificial intelligence applications in liver disease over seventeen years

Abstract

BACKGROUND

In recent years, the utilization of artificial intelligence (AI) technology has gained prominence in the field of liver disease.

AIM

To analyzes AI research in the field of liver disease, summarizes the current research status and identifies hot spots.

METHODS

We searched the Web of Science Core Collection database for all articles and reviews on hepatopathy and AI. The time spans from January 2007 to August 2023. We included 4051 studies for further collection of information, including authors, countries, institutions, publication years, keywords and references. VOS viewer, CiteSpace, R 4.3.1 and Scimago Graphica were used to visualize the results.

RESULTS

A total of 4051 articles were analyzed. China was the leading contributor, with 1568 publications, while the United States had the most international collaborations. The most productive institutions and journals were the Chinese Academy of Sciences and Frontiers in Oncology. Keywords co-occurrence analysis can be roughly summarized into four clusters: Risk prediction, diagnosis, treatment and prognosis of liver diseases. "Machine learning", "deep learning", "convolutional neural network", "CT", and "microvascular infiltration" have been popular research topics in recent years.

CONCLUSION

AI is widely applied in the risk assessment, diagnosis, treatment, and prognosis of liver diseases, with a shift from invasive to noninvasive treatment approaches.

Key Words: Artificial intelligence; Liver disease; Hepatocellular carcinoma; Liver fibrosis; Bibliometric analysis

Core Tip: This study highlights the increasing annual publications on artificial intelligence in liver disease, with applications spanning risk assessment, diagnosis, treatment, and prognosis. China leads in publication output, whereas the United States remains a dominant force in the field. High-impact journals, authors, and institutions are identified, along with trends in international collaboration. Key research hotspots include "machine learning", "deep learning", "convolutional neural networks", "CT imaging", and "microvascular infiltration".

INTRODUCTION

Liver diseases pose a global health challenge, responsible for nearly 2 million fatalities each year, and contributing to one in every 25 deaths globally[1]. The escalating prevalence of liver diseases with asymptomatic or mildly symptomatic presentations further exacerbates this issue[2], imposing an enormous economic burden on society. The high incidence and unfavorable prognosis associated with liver diseases highlight the imperative need for prompt detection and intervention. Such measures are pivotal not only in enhancing patient outcomes and quality of life but also in mitigating the financial impact on healthcare systems. In this context, the emergence of artificial intelligence (AI) technologies presents promising avenues for the early identification and management of liver diseases, paving the way for tailored precision therapies.

AI, designed to mimic human cognitive functions, is poised to revolutionize patient care quality in healthcare. Within AI, Natural Language Processing (NLP) stands out for its ability to derive meaningful clinical insights from extensive unstructured data sources, such as Electronic Health Records. NLP conducts exhaustive and unbiased analyses of these datasets, greatly aiding patient care. Machine learning (ML), another critical component of AI, has demonstrated remarkable proficiency in areas like medical imaging and disease prognosis. ML is categorized into supervised or unsupervised learning, contingent on the use of labeled training data. Deep learning (DL), an advanced branch of ML, is modeled on human brain functions, utilizing multilayered neural networks to independently interpret and learn from data’s hierarchical structures. This is particularly beneficial for tasks involving classification and feature extraction. However, DL's advanced learning abilities require substantial computational resources and entail significant expenses. Challenges include the "black box" phenomenon, where the opacity of neural network algorithms obscures the interpretability of results. Additionally, issues like data dependency and bias are prevalent, as the effectiveness of these models is largely dependent on the size and diversity of the input data. Overfitting also poses a risk, potentially limiting the model's generalizability. These challenges underscore the need for ongoing research and innovative solutions in the rapidly evolving domain of AI in healthcare.

In contemporary medical practice, a vast array of patient data has been accumulated, presenting a significant opportunity for enhanced clinical care through AI. AI's ability to process and interpret multimodal patient data promises a more holistic and accurate approach in various medical disciplines. Key areas currently employing AI include radiology, cardiology, ophthalmology, and increasingly, hepatology[3]. In hepatology, AI integration promises to improve the prognostic accuracy and diagnostic precision of hepatic disorders, thus enabling better-informed clinical decisions and positively impacting hepatic disease outcomes. Despite extensive research in hepatic disorders using AI, a comprehensive understanding in this field remains elusive. This article employs bibliometric analysis to systematically examine and visually represent scholarly articles on AI applications in hepatology from 2007 to 2023. This analysis highlights the progressive trends in AI research within the domain of hepatic diseases and anticipates future research directions.

MATERIALS AND METHODS

Data source

In this study, a comprehensive search was conducted within the Web of Science Core Collection (WoSCC) database to identify all articles and reviews pertaining to hepatopathy and AI. The timeframe for this search extended from January 2007 to August 2023. Renowned for its extensive use and acceptance in scientific and bibliometric research, the WoSCC encompasses predominant scientific journals across a multitude of disciplines globally. It provides a vast repository of academic literature, encompassing works from the early 20th century to present, thereby serving as a pivotal resource for this investigation.

Search strategy

We searched for publications related to AI in the field of liver disease within one day to ensure that there were no data updates. After reviewing the literature and organizing Medical Subject Headings terms, we identified relevant subject terms and supplemented them with free terms. The retrieval strategy is as follows: Topic = ("Liver Diseases" or "Hepatopathy" or "Liver Dysfunction" or "Liver Failure" or "Liver Neoplasm*" or "Hepatis" or "Hepatic" or "Fatty liver" or "Hepato*" or "Cirrhosis" or "Liver Abscess") AND topic = ("Artificial Intelligence" or "Computational Intelligence" or "Machine Intelligence" or "Machine Learning" or "AI" or "Computer Vision System" or "Deep Learning" or "Supervised Machine Learning" or "Unsupervised Machine Learning" or "Sentiment Analysis" or "Robotic*" or "Big Data" or "Intelligent Learning" or "Image* Segmentation" or "Knowledge Graph" or "Expert* System*" or "Semantic Segmentation" or "Feature* Learning" or "Feature* Mining" or "Feature* Extraction" or "Feature* Selection" or "Neural Network*" or "Deep Network*" or "Bayes* Network" or "Data Clustering" or "Graph Mining" or "Neural Nets Model*") AND publication year = (2007.01.01–2023.08.31). The language was limited to English, and the document types we searched for were also limited to articles and review articles.

Data analysis and visualization

For the analysis and visualization of the collected data, this study predominantly utilized tools such as VOSviewer, Citespace, and Scimago Graphica. These applications were instrumental in mapping and interpreting complex bibliometric networks. Furthermore, R software, version 4.3.1, was employed to graphically represent the evolution and trends of keyword hotspots, providing a nuanced understanding of thematic progression within the field.

In network and clustering analysis, we use the modularity Q and weighted mean silhouette S coefficients to assess the quality and effectiveness of clustering. Modularity Q measures the quality of network clustering through data partitioning, with a higher Q value indicating stronger connectivity within the same cluster. The weighted mean silhouette S assesses clustering effectiveness by reflecting the similarity of a sample to its own cluster compared with other clusters. A Q value greater than 0.3 indicates a significant clustering structure, whereas an S value above 0.7 suggests high homogeneity within the cluster[4].

Ethics statement

The data used in our study are available from public databases. Permission from the ethics committee is not needed.

RESULTS

Global trend of publications

In accordance with the outlined literature search strategy flowchart, a total of 4051 papers were extracted from the WoSCC over the past 17 years, encompassing 3530 original articles and 521 reviews (as depicted in Figure 1). A notable increase in the volume of published papers was observed from 2018-2022, signifying a growing interest in the application of AI technology in the field of liver diseases (illustrated in Figure 2). In alignment with Price's law, which posits an exponential increase in scientific documentation, the number of publications in this domain has been expanding exponentially. This trend is mathematically represented by the exponential curve equation: Y = 28.14e^0.1854×. The fitting curve, demonstrating a high degree of correlation with an R² value of 0.9132, suggests a continued upward trajectory in research output, thereby indicating that an increased number of studies in this area can be anticipated in the coming years.

Figure 1 Flowchart of study selection from the Web of Science Core Collection.

Figure 2 The number of publications on artificial intelligence research in liver disease per year from 2007-2023. The graph illustrates the exponential growth of publications from 2007-2023.

Analysis of the most productive countries

Related papers in this field have been published by authors from a collective total of 88 countries. The top 10 most productive countries contributing to the growth of AI in liver disease research activities globally are outlined in Table 1 and Figure 3A and B. China published the most studies (1568), followed by the United States (1062), Italy (318), Germany (277), Japan (258), the United Kingdom (234) and South Korea (211). China was the leading country (1568), accounting for 38.71% of all publications (4051) worldwide, but the average number of citations per paper was 13.48, ranking behind the United States (23.62). For collaboration analysis between countries, China has cooperated with many countries, most importantly with the United States and the United Kingdom (Figure 3C and D). Additionally, VOSviewer was employed to construct a network map illustrating the collaborative relationships among countries (as shown in Figure 3E). This analysis encompassed 30 countries, each with a minimum publication count exceeding 32. In this map, the lines connecting various nodes signify collaborative interactions between countries, with the line thickness indicating the intensity of these collaborations, quantified as the total link strength (TLS). The co-authorship visualization highlighted that the United States, China, Italy, the United Kingdom, and Germany were the top five countries in terms of TLS. The map of the cooperation network between countries shows that the United States has more connections with other countries, including China, the United Kingdom, Germany, and Italy. This pattern underscores the global scope of research attention in the application of AI in liver disease studies.

Figure 3 Visualization of publications and cooperation between countries. A: A world map color-coded by the number of publications per country; B: A bar graph showing the number of publications from the top 10 most productive countries from 2007-2023; C-E: The map of international cooperation has been produced via a website (https://charticulator.com), Scimago Graphica, and VOSviewer.

Table 1 The top 10 productive countries related to artificial intelligence for liver disease.

Country	Counts	Ranking on the basis of total documents	Total citations	Ranking on the basis of total citations	Average citation	Ranking on the basis of total link strength	Link strength
China	1568	1	21137	2	13.48	2	678
United States	1062	2	25084	1	23.62	1	1034
Italy	318	3	6629	3	20.85	3	605
Germany	277	4	6377	4	23.02	5	539
Japan	258	5	5716	5	22.16	7	374
United Kingdom	234	6	5312	6	22.70	4	559
South Korea	211	7	3935	8	18.65	9	339
France	175	8	4423	7	25.27	6	503
India	165	9	3245	9	19.67	18	167
Canada	118	10	3066	11	25.98	11	261

Contributions of top institutions

This study identified a total of 4957 institutions contributing to the publication of research on AI in liver disease. As summarized in Table 2, the top 10 institutions with the highest contributions are predominantly based in China, with the exception of the University of Pittsburgh from the United States, which ranks seventh. Notably, the University of Pittsburgh also has the distinction of having the highest average number of citations. The pattern of institutional collaboration is presented in Figure 4A. In terms of publication volume, the Chinese Academy of Sciences led with 104 publications, closely followed by Fudan University (96), Zhejiang University (93), and Sun Yat-sen University (92). Figure 4A shows that among the institutions with over 15 publications, the Chinese Academy of Sciences (TLS = 263), Fudan University (210), and Sun Yat-sen University (179) have the highest TLS. The centrality of organizations in this research area was analyzed via the CiteSpace platform. Figure 4B, generated by CiteSpace, visualizes the cooperation between organizations. Organizations with a purple outer ring in the visualization represent a centrality greater than 0.1, indicating a central role in the network. Notably, the National University of Singapore emerged as a central player (centrality 0.12), highlighting its significant position in the network of institutional collaboration in AI research on liver disease.

Figure 4 Network visualization of institutions involved in artificial intelligence research on liver disease. A: Produced by VOSviewer; B: Produced by CiteSpace.

Table 2 The top ten institutions by volume of the application of artificial intelligence to liver disease.

Rank	Country	Counts	Institution	Citation	Mean citation
1	China	104	Chinese Academy of Sciences	2088	20.08
2	China	96	Fudan University	1465	15.26
3	China	93	Zhejiang University	1366	14.69
4	China	92	Sun Yat-Sen University	1833	19.92
5	China	64	Shanghai Jiao Tong University	732	11.44
6	China	63	Sichuan University	606	9.62
7	United States	56	University of Pittsburgh	1815	32.41
8	China	53	Capital Medical University	735	13.87
9	China	52	Stanford University	1308	25.15
10	China	47	Chinese People’s Liberation Army General Hospital	858	18.26

Analysis of authors

The study included a comprehensive cohort of 129243 authors. For a more focused analysis in VOSviewer, authors who had published more than five articles were specifically included, facilitating the generation of a coauthorship visualization graph, as depicted in Figure 5. Table 3 delineates the ten most prolific authors in terms of publication volume. The leaders of this group are Sucandy Iswanto, Ross Sharona, and Rosemurgy Alexander, who contributed 32, 25, and 23 papers, respectively.

Figure 5 Visualization map of the coauthorship analysis of authors carried out on VOSviewer.

Table 3 The top ten most productive authors on artificial intelligence in liver disease.

Rank	Author	Counts	Total citations	Mean citations
1	Sucandy Iswanto	32	501	15.66
2	Ross Sharona	25	176	7.04
3	Rosemurgy Alexander	23	442	19.22
4	Wang Wei	19	23	1.21
5	Tian Jie	17	508	29.88
6	Goh Brian K. P.	17	177	10.41
7	Liu Rong	17	225	13.24
8	Pawlik Timothy M.	17	218	12.82
9	Song Bin	16	131	8.19
10	Di Benedetto Fabrizio	15	546	36.40

Contributions of the top journals

In this study, a total of 1116 journals were identified as having published articles relevant to the research topic. Among these, 189 journals published no fewer than five articles each. The quality of a journal is often assessed via metrics such as the impact factor (IF, JCR category) and the number of citations received.

Table 4 lists the top ten journals by publication count, with Frontiers in Oncology (89 articles, IF = 4.7, Q2), Scientific Reports (82 articles, IF = 4.6, Q2), and Asian Journal of Surgery (77 articles, IF = 3.5, Q1) occupying the top three positions. However, in terms of average citations, the leading journals are IEEE Transactions on Medical Imaging (139.60), Annals of Surgery (132.15), Proceedings of the National Academy of Sciences of the United States of America (116.20), and Neurocomputing (105.38). This suggests that the articles published in these journals are of high quality. Figure 6, created via CiteSpace software, presents a dual map of journals, illustrating the relationship between cited and citing journals. This visual representation reveals four main citation pathways. The cited articles predominantly fall into two fields: (1) Medical, clinical, and (2) Molecular, biological, and immunology. However, the citing papers are primarily concentrated in three fields: (1) Molecular, Biology, Genetics; (2) Health, Nursing, Medicine; and (3) Dermatology, Dentistry, Surgery. This delineation provides insight into the interdisciplinary nature and citation dynamics within the scope of the study. The concentration of citing papers in the fields of dermatology and dentistry may be attributed to interdisciplinary collaborations or overlapping methodologies.

Figure 6 A dual-map overlap of journals on artificial intelligence research in liver disease was carried out via CiteSpace.

Table 4 The top 10 journals related to artificial intelligence in the field of liver disease.

Rank	Journal title	Countries	Count	Impact factor (2023)	JCR (2023)	Total citations
1	Frontiers in Oncology	Switzerland	89	4.7	Q2	655
2	Scientific Reports	England	82	4.6	Q2	1000
3	Asian Journal of Surgery	China	77	3.5	Q1	553
4	European Radiology	Germany	73	5.9	Q1	1672
5	Surgical Endoscopy and Other Interventional Techniques	United States	73	3.1	Q1	1763
6	Cancers	Switzerland	69	5.2	Q2	317
7	Plos One	United States	67	3.7	Q2	1321
8	World Journal of Gastroenterology	China	55	4.3	Q2	950
9	Diagnostics	Poland	51	-	-	159
10	Medical Physics	United States	42	3.8	Q2	717

IF: Impact factor; JCR: Journal Citation Reports.

Analysis of top cited references and co-citation references

In this research, a total of 4051 papers were analyzed, among which 303 were cited more than 50 times. Table 5 presents the top 10 most cited articles pertaining to the application of AI in liver disease. These publications are indicative of foundational works in the integration of AI within this specific medical domain. The work by Bray et al[5] has the most citations, with 148 citations, followed by Yasaka et al[6] and the European Association for the Study of the Liver, with 128 and 117 citations, respectively. Figure 7A provides a visual representation of the cocitation network of references, demonstrating effective clustering, as evidenced by high modularity Q and weighted mean silhouette S values of 0.9015 and 0.9704, respectively. Additionally, Figure 7B offers a timeline view of cocited articles, highlighting the evolution of research hotspots over time. For clarity, only the top 8 Largest clusters are depicted in this figure. The four largest clusters identified are "hepatocellular carcinoma", "robotic liver surgery", "liver fibrosis", and "nonalcoholic steatohepatitis", highlighting the key areas of focus in AI-related liver disease research.

Figure 7 CiteSpace visualization map of cocited references. A: Cluster view; B: Timeline view.

Table 5 The top ten most cited articles related to artificial intelligence in the field of liver disease.

Title	Journals	First author	Year	Citations
Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries	CA Cancer J Clin	Freddie Bray	2018	148
Deep Learning with Convolutional Neural Network for Differentiation of Liver Masses at Dynamic Contrast-enhanced CT: A Preliminary Study	Radiology	Koichiro Yasaka	2018	128
EASL Clinical Practice Guidelines: Management of hepatocellular carcinoma	J Hepatol	European Association for the Study of the Liver	2018	117
Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries	CA Cancer J Clin	Hyuna Sung	2021	112
Diagnosis, Staging, and Management of Hepatocellular Carcinoma: 2018 Practice Guidance by the American Association for the Study of Liver Diseases	Hepatology	Jorge A Marrero	2018	109
AASLD guidelines for the treatment of hepatocellular carcinoma	Hepatology	Julie K Heimbach	2018	91
Radiomics: Images Are More than Pictures, They Are Data	Radiology	Robert J Gillies	2016	89
Radiomic analysis of contrast-enhanced CT predicts microvascular invasion and outcome in hepatocellular carcinoma	J Hepatol	Xun Xu	2019	88
Deep learning Radiomics of shear wave elastography significantly improved diagnostic performance for assessing liver fibrosis in chronic hepatitis B: A prospective multicenter study	Gut	Kun Wang	2019	85
A survey on deep learning in medical image analysis	Med Image Anal	Geert Litjens	2017	78

Analysis of keyword co-occurrences

Keyword co-occurrence analysis can help researchers find the key topics and topic structure in the research field to better understand their characteristics. However, keyword bursts refer to the frequent occurrence of a certain keyword in a certain period, through which the emerging hot spots and major events in the research field can be identified. A total of 13667 author keywords were included in our study, of which 138 appeared more than 35 times. Table 6 lists the top 30 keywords that appear most frequently. The most frequent keyword was “hepatocellular carcinoma”, with 1023 keywords, followed by “machine learning” (551 times) and “deep learning” (393 times). We use VoSviewer to produce a network visualization map (Figure 8A and B) and an overlay visualization map (Figure 8C) of keywords related to AI in liver disease research. In Figure 8A and B, the high-frequency keywords (frequency not less than 35) included in this study can be divided into four clusters, and each color indicates a cluster that represents diverse research tendencies. There are four clusters in total: The red cluster focuses on risk prediction of liver disease, such as keywords "machine learning", "model", "risk", "artificial neural network", "fatty liver disease", and "obesity"; the green cluster tends to be associated with liver disease imaging and diagnosis, such as keywords "image", "ultrasonography", "accuracy", "classification", "artificial intelligence", "convolutional neural network", and "prediction"; the blue cluster focuses on the treatment of liver diseases, especially hepatocellular carcinoma (HCC), such as keywords such as "hepatocellular carcinoma", "hepatectomy", "laparoscopic surgery", "robotic surgery" and "chemotherapy"; and the yellow cluster focuses on the prognosis of liver diseases. Since the graph includes keywords with a frequency greater than 50 times, the yellow cluster includes only three keywords: "prognosis", "nomogram" and "microvascular invasion". Figure 8C and D show the change in keywords over time, and the yellow nodes in Figure 8C represent emergent keywords, indicating research hotspots in recent years. The "machine learning", "deep learning", "convolutional neural network", "CT", and "microvascular infiltration" methods receive more attention.

Figure 8 Visualization analysis of keyword co-occurrence. A and B: Keyword co-occurrence and clustering via VOSviewer; C: Overlay visualization of keywords; D: Heatmap of annual keyword frequency.

Table 6 The top 30 keywords with the highest frequency of application of artificial intelligence to liver disease.

Rank	Keywords	Count	Rank	Keywords	Count	Rank	Keywords	Count
1	Hepatocellular carcinoma	1023	11	Survival	206	21	CT	145
2	Machine learning	551	12	Model	202	22	System	144
3	Deep learning	393	13	Resection	200	23	Neural network	139
4	Cancer	377	14	Outcome	195	24	Experience	133
5	Diagnosis	310	15	Disease	190	25	Robotic surgery	132
6	Classification	294	16	Expression	190	26	Liver resection	124
7	Artificial intelligence	291	17	Liver	175	27	Identification	122
8	Surgery	235	18	Management	169	28	Computed tomography	120
9	Prediction	219	19	Convolutional neural network	167	29	Fatty liver disease	120
10	Risk	210	20	Cirrhosis	163	30	Cell	116

DISCUSSION

Global research status and trends

Over the past two decades, research on AI in liver diseases has exhibited exponential growth. Given China's vast population and the increasing prevalence of liver diseases, its contributions have notable global health implications[7,8]. However, the findings indicate that while the volume of publications from China is substantial, their global impact remains relatively limited, pointing to a need for more in-depth and high-quality research outputs. Several factors might contribute to this phenomenon: (1) In recent years, the Chinese government has robustly supported the development of AI applications, leading to a significant increase in research institutions and publications within China[9,10]; (2) The effectiveness of AI outputs has been closely linked to the quality of input data[11]. The advancement of AI technology hinges on powerful computing capabilities and sophisticated algorithms, areas in which China's development is still progressing; and (3) Language barriers may also play a role, potentially limiting the international citation of Chinese literature. These factors collectively suggest areas for improvement and development to enhance the global influence of China's research in AI applications for liver diseases.

China and the United States rank highest in both publication volume (China: 1568; United States: 1062) and total citations (China: 21137; United States: 25084). They also act as core countries in the international collaboration network, further underscoring their pivotal roles in advancing this field. Additionally, Canada, France, Germany, the United Kingdom, and Japan exhibit outstanding performance in terms of average citations. Specifically, Canada has an average citation of 25.98, France 25.27, Germany 23.02, the United Kingdom 22.70, and Japan 22.16. This reflects these countries' significant contributions to high-impact research, particularly through active participation in international collaborations and scientific innovation.

Keyword analysis

Cluster analysis of the literature revealed a growing trend in the publication of articles focused on liver cancer[12], robotic liver surgery, liver fibrosis, and nonalcoholic fatty liver disease (NAFLD) in recent years. These topics can be broadly categorized into four clusters, representing key areas in liver disease research: Risk prediction (represented by the red cluster), diagnosis (green cluster), treatment (blue cluster), and prognosis (yellow cluster). As depicted in Figure 8C, initial studies predominantly concentrated on the diagnosis and surgical treatment of liver diseases. Recent trends, however, indicate a shift toward topics such as "machine learning", "deep learning", "convolutional neural networks", "CT imaging", and "microvascular infiltration". This shift in focus points to an evolving research landscape where the application of AI in liver diseases is transitioning from traditional invasive methods to more advanced, noninvasive techniques. This progression underscores the dynamic nature of AI integration in medical research, particularly in the context of liver disease management.

Application of AI in the risk prediction of liver diseases: The etiology of liver diseases, including HCC and NAFLD, is profoundly influenced by individual and temporal factors, posing significant challenges in the development of predictive models for these diseases. Traditional regression analyses often oversimplify these models by excluding variables deemed "nonsignificant", which compromises their accuracy and generalizability. In contrast, AI techniques, which leverage electronic medical records and extensive databases, have shown potential in identifying populations at heightened risk for conditions such as NAFLD and, further, in stratifying this risk with greater precision[13,14]. Risk scoring systems have been developed to evaluate the likelihood of developing HCC in patients with chronic liver disease, and these systems have demonstrated acceptable performance levels[15,16]. However, the clinical application of these risk scores remains limited, primarily due to the restricted size and diversity of the training datasets. This limitation hampers the ability of the models to be generalized externally, thus affecting their practical utility in diverse clinical settings.

Application of AI in the diagnosis of liver diseases: AI integration with various diagnostic methods, such as radiomics, histopathology, and molecular biology, is poised to significantly increase diagnostic accuracy in liver disease[3]. As the largest solid organ in the human body, the liver primarily depends on imaging techniques for initial disease diagnosis. However, imaging results are often qualitative, such as those from ultrasound, or semiquantitative, such as those from CT scans, leading to potential interpretative discrepancies among clinicians. Furthermore, the repetitive task of image analysis can induce physician fatigue[17] and increase the likelihood of errors. Radiomics texture analysis, which is capable of rapidly extracting subtle imaging features that may elude the human eye[18], offers a more comprehensive and objective interpretation, thus aiding in enhancing diagnostic efficiency.

Although radiological examinations are often sufficient for diagnosing HCC, bypassing the need for tissue biopsy, histopathological examination remains the definitive standard for many liver conditions, including the characterization of focal liver lesions, as well as the grading and staging of autoimmune hepatitis and nonalcoholic steatohepatitis. Research indicates that histology reveals extensive information about potential molecular changes and disease prognosis[19-21], providing deeper insights into the pathophysiology and prognosis of liver diseases. Recent advances in sequencing technologies have led to the use of single-cell RNA sequencing (scRNA-seq), which offers a detailed view of the cellular heterogeneity in liver diseases, whereas bulk RNA sequencing provides an overview of the overall pathological state. The synergistic application of these methodologies has enhanced our understanding of the molecular basis of liver diseases and aided in identifying potential therapeutic targets[22]. Currently, single-cell technologies are still evolving, and AI can play a pivotal role in addressing challenges such as data quality, the loss of low-expression data during sequencing, and data noise[23,24]. However, the absence of standardized data formats and norms presents significant challenges in data sharing and comparisons in single-cell research.

Application of AI in the treatment of liver diseases: The application of AI in liver disease treatment primarily focuses on optimizing treatment plan selection and enhancing intraoperative visualization. A critical issue in liver cancer therapy involves accurately predicting the most effective treatment approach or combination for each patient[25], thereby facilitating prompt and informed clinical decisions. The IMBRAVE-150 trial[26] demonstrated that, compared with sorafenib, a combination therapy of atezolizumab and bevacizumab significantly improved survival rates in HCC patients. These findings underscore the potential role of immune checkpoint inhibitors (ICIs) in cancer therapy. However, given that not all HCC patients benefit from ICIs and considering their substantial side effects and costs, the AI-based prediction of treatment response becomes vital[3].

Research has shown that intraoperative complications often arise from errors in visual anatomical assessments[27,28]. DL shows immense promise in computer vision tasks, but its application in real-time surgical guidance is challenging because of the variable intraoperative environment of liver surgeries[29]. Madani et al[29] developed a DL model capable of identifying anatomical landmarks and distinguishing between safe and risky zones during laparoscopic cholecystectomy. This model has demonstrated high efficacy, illustrating the potential of AI in providing real-time intraoperative guidance. Liver transplantation is a critical life-saving procedure for patients with end-stage liver disease. Given the substantial gap between organ availability and demand, AI applications can assist in evaluating transplant risks and in matching donors with recipients [e.g., using Model for End-Stage Liver Disease (MELD) and MELD-Na models] before transplantation. This can help reduce mortality on waiting lists. Posttransplantation, AI can be instrumental in identifying the risks of recurrence and complications, thereby assisting in the management of transplant recipients[30]. However, the integration of these AI technologies in clinical practice necessitates thorough clinical validation and careful consideration of ethical implications.

Application of AI in the prognosis of liver diseases: Microvascular infiltration (MVI), a crucial histological characteristic in the treatment and prognosis of liver cancer, is closely linked with postoperative recurrence of the disease. Preoperative identification of MVI status can significantly aid physicians in making more informed medical decisions. Traditionally, the detection of MVI relies on postoperative tissue biopsies, which inherently introduces a delay. Additionally, MVI often occurs interspersed within liver tissues adjacent to the tumor[31], necessitating multipoint sampling around the tumor area to ensure accurate detection. However, such extensive sampling is challenging in clinical practice because of factors such as the need to preserve liver function[32]. With advancements in AI, there is now burgeoning potential for effectively evaluating MVI status via AI algorithms applied to histological whole-slide images of HCC. This technological progress holds promise for enhancing the clinical management of HCC patients by providing a more timely and accurate assessment of MVI[33].

AI: The challenge

AI has promising prospects in the field of hepatology, but many challenges remain in clinical practice. First, AI holds significant potential in hepatology, yet its implementation in clinical settings is met with several challenges. First, the lack of standardized protocols in AI-mediated medical data analysis is a concern, with the majority of studies being retrospective. This underscores the need for more prospective research to refine and validate AI models. Second, simplistic designs in algorithmic models can inadvertently create healthcare disparities. An example is the MELD, which is used for liver transplant prioritization. MELD is based on three laboratory parameters: The bilirubin level, international normalized ratio, and creatinine level. The issue arises with creatinine levels, which are typically lower in women, possibly owing to lower muscle mass, leading to an underestimation of renal insufficiency in female patients. Additionally, the MELD system awards extra points for HCC, disproportionately benefiting males who have a higher incidence of HCC[34]. This can disadvantage female patients in terms of transplant allocation, potentially leading to higher mortality rates among women on the transplant list[35]. ML, particularly DL, depends heavily on the size and diversity of the training dataset, with the risk of overfitting and biased outcomes if the data are not representative[36]. To mitigate this, the promotion of data sharing is essential, but it must be balanced with patient privacy concerns. Furthermore, the "black box" nature of AI, especially in DL models, presents a barrier to trust and acceptance among clinicians and patients due to the lack of transparency in the decision-making process. To address this issue, there is a growing focus on developing methods to improve the interpretability of AI models. Techniques such as explainable AI are being explored to demystify the decision-making processes within these models.

Limitations

This study has certain limitations: (1) We included only English literature in the WoSCC, and the document types included only articles and reviews, which results in a greater probability of missing literature; (2) The search time of our study was up to 2023.08.31, and the time delay of new studies may result in the omission of emerging research hotspots; and (3) Bibliometric analyses may be subject to biases, such as variations in citation practices across different academic fields or countries, which can lead to certain papers being cited more frequently while others are overlooked. Additionally, newly published papers may not have accumulated sufficient citations, resulting in emerging research not being adequately represented.

CONCLUSION

In conclusion, AI has been extensively applied in various aspects of liver disease management, including risk assessment, diagnosis, treatment, and prognosis. While a significant volume of the literature originates from China, there is a recognized need to increase the quality of these publications. The United States continues to maintain its leadership in this research domain. Emerging trends and future research directions are predominantly focused on areas such as "machine learning", "deep learning", "convolutional neural networks", "CT imaging", and "microvascular infiltration". These trends suggest an increasing emphasis on leveraging AI for the prognostic analysis of liver diseases. Moreover, there has been a noticeable shift in the application of AI within this field, moving from invasive methodologies to more noninvasive approaches. This transition points to the evolving nature of AI technology and its growing integration into less intrusive yet effective medical procedures and analyses.