- Research
- Open access
- Published:
Image-based AI diagnostic performance for fatty liver: a systematic review and meta-analysis
BMC Medical Imaging volume 23, Article number: 208 (2023)
Abstract
Background
The gold standard to diagnose fatty liver is pathology. Recently, image-based artificial intelligence (AI) has been found to have high diagnostic performance. We systematically reviewed studies of image-based AI in the diagnosis of fatty liver.
Methods
We searched the Cochrane Library, Pubmed, Embase and assessed the quality of included studies by QUADAS-AI. The pooled sensitivity, specificity, negative likelihood ratio (NLR), positive likelihood ratio (PLR), and diagnostic odds ratio (DOR) were calculated using a random effects model. Summary receiver operating characteristic curves (SROC) were generated to identify the diagnostic accuracy of AI models.
Results
15 studies were selected in our meta-analysis. Pooled sensitivity and specificity were 92% (95% CI: 90–93%) and 94% (95% CI: 93–96%), PLR and NLR were 12.67 (95% CI: 7.65–20.98) and 0.09 (95% CI: 0.06–0.13), DOR was 182.36 (95% CI: 94.85-350.61). After subgroup analysis by AI algorithm (conventional machine learning/deep learning), region, reference (US, MRI or pathology), imaging techniques (MRI or US) and transfer learning, the model also demonstrated acceptable diagnostic efficacy.
Conclusion
AI has satisfactory performance in the diagnosis of fatty liver by medical imaging. The integration of AI into imaging devices may produce effective diagnostic tools, but more high-quality studies are needed for further evaluation.
Background
Fatty liver disease has become more and more prevalent in recent years [1], making it the most common chronic liver disease in the world. Fatty liver can lead to steatohepatitis, liver fibrosis, cirrhosis, and even hepatocellular carcinoma, early detection and treatment may stop or even reverse the progression of fatty liver [2]. The best reference for diagnosis and classification of hepatic steatosis is the liver biopsy [3]. Nevertheless, the high cost [4], sampling errors [5, 6], and procedure-related morbidity and mortality [7] make it unsuitable for screening. Therefore, it is urgent and necessary to develop non-invasive diagnostic tools to assess hepatic steatosis.
Imaging is a useful tool to assist decisions of diagnosis, staging, and treatment in clinical practice. Currently, the main diagnostic modalities by medical imaging for fatty liver include magnetic resonance imaging (MRI), ultrasound (US), and computed tomography (CT). Conventional US is cheap, safe, and non-invasive, so it is the most commonly used modality for clinical screening [8]. But the diagnostic accuracy in the US is largely dependent on personal judgment which may be susceptible to many factors. CT can effectively detect fatty liver without the influence of abdominal fat. But it is radioactive and expensive, besides, the classification of fatty liver by CT value may be too rough. MRI has high soft tissue resolution and can quantify intrahepatic fat at the molecular level, so it is the main modality for the non-invasive quantification of hepatic steatosis [9]. However, the high cost and difficult operation may limit its clinical application. In institutions with limited medical resources, the lack of imaging equipment and experts will make it challenging to obtain the accurate and immediate diagnosis through medical imaging [10].
Artificial intelligence(AI) has made significant advances since the 21st century, especially in medical imaging diagnosis [11], such as conventional machine learning(ML) and deep learning(DL). Concerning the application of AI in medical imaging, a large number of quantitative features can be extracted from radiological images using sophisticated image processing techniques, which are subsequently analyzed by traditional biostatistical or AI models to diagnose or assess therapeutic responses. Several AI-assisted diagnostic models have been developed for fatty liver, such as Han et al. [12] who developed a classifier for the diagnosis of nonalcoholic fatty liver disease(NAFLD), obtaining 97% for sensitivity and 94% for specificity. The model was established by DL using US radio frequency (RF) data with reference to MRI-derived proton density fat fraction (PDFF). Many scholars are trying to improve the diagnostic efficacy of AI models by optimizing image quality, expanding sample size, and modifying algorithms.
To date, little meta-analysis has been conducted to evaluate the diagnostic performance of image-based AI. The study aimed to perform a systematic review and meta-analysis to assess the performance of image-based AI in the diagnosis of fatty liver.
Methods
Protocol registration and study design
The study was registered in the PROSPERO(CRD42023388607). The meta-analysis took the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) guideline [13] as the reference.
Search strategy
We searched Embase, Pubmed, and Cochrane library for studies of image-based AI in fatty liver until December 24, 2022. The search terms were as follows: “artificial intelligence”, “deep learning”, “machine learning”, “fatty liver”, “NAFLD”, “non-alcoholic fatty liver disease”, “steatohepatitis”, “metabolic dysfunction-associated fatty liver disease” and “diagnosis, computer-assisted”. The detailed search strategies for each database were summarised in Table S1.
Inclusion and exclusion criteria
We included all articles that used AI in the imaging diagnosis of hepatic steatosis. The inclusion criteria: (1) participants underwent fatty liver-related imaging; (2) references were accurately described. The exclusion criteria: (1) duplicate publications; (2) non-English articles; (3) reviews, meta-analyses, comments, editorials, guidelines, and conference abstracts; (4) non-human samples; (5) pathological images, combined with non-image information, without AI models; (6) studies without enough information to calculate true positive (TP), true negative (TN), false positive (FP), and false negative (FN) values. The titles and abstracts were independently screened according to the eligibility criteria by two reviewers (L-YD and Z-Q), and subsequently downloaded and reviewed the full text.
Data extraction
Two authors (L-YD and Z-Q) conducted the data extraction independently. Any disagreements about the data were determined with the third author(Y-XJ). Data extraction included authors, years, countries, study design, eligibility criteria, age, sample size, data source and range, imaging technique, reference, AI algorithm, and TP, FP, TN, FN values, which were used to calculate sensitivity and specificity. For studies that developed more than one AI model, we selected the one with the best overall performance for analysis.
Quality assessment
Two independent evaluators (L-YD and Z-Q) assessed the quality of all selected studies by the Quality Assessment of Diagnostic Accuracy Studies-AI (QUADAS-AI) criteria [14]. The guideline includes four domains in the risk of bias and three domains of applicability (Table S2). The new tool, a combination of QUADAS-2 [15] and QUADAS-C [16], was specifically designed to assess the risk and suitability of bias in AI associated studies. All disagreements were discussed with a third collaborator (Y-XJ).
Statistical analysis
The quality of all selected studies was assessed by RevMan using QUADAS-AI, the risk of publication bias was assessed by Stata software (version 17.0) and all other statistical analysis was conducted in Meta-disc (version 1.4). Spearman’s correlation coefficient between the log of sensitivity and the log of (1-specificity) was calculated to test the threshold effects, and heterogeneity was tested using the I2 statistic. A random effects model was used to calculate pooled sensitivities, specificities, negative likelihood ratios (NLR), positive likelihood ratios (PLR), diagnostic odds ratios (DOR), and their 95% confidence intervals(CI) based on crude values of TP, TN, FP and FN values for each study. Summary receiver operating characteristic curves (SROC) were fitted to assess the accuracy of the AI models. The low, medium, and high accuracy were defined as the area under the curve (AUC) values of 0.5–0.7, 0.7–0.9, and 0.9-1 respectively [17]. Subgroup analyses were then performed: (1) AI algorithm (conventional ML or DL); (2) region; (3) whether transfer learning was performed; (4) reference (US, MRI or pathology); (5)imaging techniques (conventional US, elastography or MRI). The risk of publication bias was assessed by Deeks funnel plots. Fagan plots were drawn to calculate the pre-test and post-test probabilities to evaluate the clinical value. P-values < 0.05 were then considered statistically significant.
Results
Study selection
The flow of searching and selecting articles was shown in Fig. 1. Finally, 15 articles [12, 18,19,20,21,22,23,24,25,26,27,28,29,30,31] were taken into the quantitative analysis. The description of all selected studies was presented in Table 1.
Quality assessment
The detailed results of quality assessments of included studies were presented in Figure S1. The risk of bias was shown in more than half of the studies for patient selection (n = 8) and index test (n = 15) because of the lack of detailed descriptions of included patients and appropriate external validation.
Overall performance of AI models
The detailed information on contingency tables and performance of AI models from 15 included studies was shown in Table S3. The meta-analysis indicated that image-based AI models were effective to diagnose liver steatosis with pooled sensitivity and specificity of 92% (95% CI: 90–93%) and 94% (95% CI: 93–96%), PLR and NLR of 12.67 (95% CI: 7.65–20.98) and 0.09 (95% CI: 0.06–0.13), DOR of 182.36 (95% CI: 94.85-350.61), and SROC of 0.98. (Fig. 2).
Subgroup meta-analysis
We performed the subgroup analysis of AI algorithm, region, reference, imaging technique and transfer learning. In AI algorithm, the pooled sensitivity and specificity of 9 conventional ML studies were 94% (95% CI: 91–96%) and 91% (95% CI: 87–94%), and were 91% (95% CI: 88–93%) and 97% (95% CI: 95–98%) in 6 DL studies. For different regions, 6 studies were conducted in Asia with pooled sensitivity and specificity of 96% (95% CI: 93–98%) and 92% (95% CI: 87–96%), 9 studies were in Europe and America with pooled sensitivity and specificity of 90% (95% CI: 88–92%) and 95% (95% CI: 93–97%). For different references, the sensitivities of US, pathology and MRI were 92%, 91% and 92%, and the specificities were 97%, 88%, and 90% respectively. For different imaging techniques, the sensitivities of conventional US, elastography and MRI were 94%, 89% and 93%, and the specificities were 94%, 96%, and 81%. Two studies employed transfer learning with pooled sensitivity and specificity of 88% (95% CI: 85–91%) and 98% (95% CI: 97–99%), 13 studies did not perform transfer learning with pooled sensitivity and specificity of 95% (95% CI: 92–96%) and 92% (95% CI: 89–94%). The details of subgroup analysis were shown in Table S4.
Heterogeneity analysis
There was substantial heterogeneity between the included studies, with I2 = 60.1% (p = 0.001) for sensitivity, I2 = 63.8% (p < 0.001) for specificity, I2 = 65.8% (p < 0.001) for PLR, I2 = 41.0% (p = 0.049) for NLR, I2 = 47.2% (p = 0.022) for DOR. The Spearman correlation coefficient was − 0.148 (p = 0.598), indicating that there was no threshold effect. And the heterogeneity was reduced after subgroup analysis which were presented in Table S4.
Clinical value and publication bias
The post-test probability of image-based AI for the diagnosis of hepatic steatosis was 94%, much higher than the pre-test probability (50%), indicating that image-based AI is valid for the diagnosis of hepatic steatosis (Fig. 3a). And the Deeks funnel plot revealed no obvious publication bias of included studies (P = 0.38) (Fig. 3b).
Discussion
AI has been widely used in medical imaging in recent years, so more and more AI models have been established to diagnose various liver diseases [32, 33]. We conducted an extensive literature search in medical databases, which was carefully screened and critically assessed by QUADAS-AI. Ultimately, we found that AI models performed well in identifying liver steatosis by medical imaging.
AI aims to simulate, extend and expand human intelligence [34]. Conventional ML is the method to achieve AI, which can use features extracted from various kinds of data to build prediction models through different algorithms. However, it requires manual extraction of features [35] and the ability of conventional ML to learn from the data was limited [36]. DL is the advanced classification of conventional ML which can utilize multiple layers of deep neural networks for a deeper understanding of the data [37]. However, DL is prone to overfitting and usually requires more data [38]. Our subgroup analysis of the different algorithms showed that the sensitivity was higher in conventional ML, but the specificity, PLR, DOR and SROC were higher in DL. The results revealed the potential advantages of DL in the image-based diagnosis of liver steatosis.
Machine learning is commonly employed in biomedical fields. However, due to insufficient labeled data, the application of advanced machine learning algorithms in clinical settings is limited. Collecting labeled data is time-consuming, energy-draining, and requires professional expertise. To address this problem, transfer learning can transfer the acquired knowledge and models from one related task to another, leading to enhanced performance and generalization of the target task [39]. For instance, a recent study utilized transfer learning to diagnose corona virus disease (COVID-19) automatically through CT images with a remarkable accuracy of 99.60% [40]. In our subgroup analysis, we found that transfer learning led to higher specificity, PLR, and DOR, which highlighted the significance of transfer learning in image-based AI diagnosis of hepatic steatosis. However, only two studies exploited transfer learning, further studies are needed to confirm its effectiveness.
The gold standard for the diagnosis of hepatic steatosis is pathology, but there are diagnostic errors in the liver biopsy due to the limitation of sampling. The EASL Clinical Practice Guideline [41] demonstrated that the MRI-PDFF was the most accurate non-invasive method for detecting and quantifying steatosis. So the articles which used experts diagnostic US or MRI-PDFF as references were also selected in our study. We further conducted the subgroup analysis of different reference standards. The results showed a higher sensitivity and lower specificity in studies taking pathology as the reference compared to US and MRI. This result indirectly demonstrated the previously mentioned limitations of pathology in terms of sampling error. Only part of the liver tissues was taken for pathological examination. When steatosis was slight or focal, false negatives were likely to occur, resulting in low specificity of AI model diagnosis. Therefore, there is an urgent need for image-based AI models with high diagnostic efficacy, which can be integrated into imaging examination equipment.
For the imaging technique, conventional US, elastography and MRI were included in the selected citations. The subgroup analysis of different imaging techniques showed that the sensitivity and DOR was higher in conventional US than elastography and MRI, which demonstrated that AI seems to be more useful in the conventional US. However, the researches on elastography and MRI were too limited, and the source of the data were different. In the future, further researches are needed to explore the AI-assistant efficacy of different imaging techniques.
Additionally, in our subgroup analysis of different regions, we found the sensitivity, DOR and SROC were higher in Asia, the specificity and PLR were higher in Europe and America, which suggested that the regions of included population may influence the diagnostic efficacy of AI for the diagnosis of hepatic steatosis. Most of the studies we included were based on US images, which are susceptible to body size and visceral fat [42]. Westerners are fatter than Asians with greater differences between populations, which may affect the accuracy of AI diagnosis. In the future, an accurate description such as body size and visceral fat of included populations will be needed, so that we can explore the potential influences on the diagnostic efficacy of AI for hepatic steatosis.
There are some advantages in our study. Our study shows the high efficacy of image-based AI in diagnosing hepatic steatosis without publication bias and may provide a reference for future clinical practice. Compared with the previous systematic review of AI-assistant in NAFLD [43], our study mainly explore the diagnostic performance of image-based AI for liver steatosis rather than fibrosis. The number of cited papers (15 citations) was increased and the subgroup analysis of different imaging techniques, AI algorithm, regions and so on may be helpful for future researches. In addition, we employed the new tool QUADAS-AI involve in AI-specific methodology in our study. In the past, the most frequently utilized quality assessment tool for the diagnostic meta-analysis was QUADAS-2. However, it does not involve in AI-specific methodology, such as generalizability and diversity in patient selection, development of training, validation and testing datasets, as well as definition and evaluation of an appropriate reference standard [44]. This new tool QUADAS-AI14 is an AI-specific extension of QUADAS-2 and QUADAS-C, includes four domains in the risk of bias and three domains in applicability concerns, which is more comprehensive and suitable for AI associated studies. Some studies [45,46,47] related to AI models have also employed this new tool.
However, our study has some limitations: firstly, most of the studies were retrospective and did not clearly describe the participants, making it difficult to control many confounding factors. Secondly, none of the included studies underwent suitable external validation, so whether the model can be applied to other populations requires further validation. Finally, there was heterogeneity in our meta-analysis, but no significant threshold effects were found according to Spearman’s correlation coefficient and the heterogeneity was reduced in the subgroups which might be the potential resources of the heterogeneity. In the future, we hope that more prospective AI studies with external validation based on large sample sizes can accurately assess the performance of image-based AI in diagnosing liver steatosis.
Conclusion
This meta-analysis suggested that AI had vast potential for image-based diagnosis of hepatic steatosis. The integration of AI into imaging devices may produce effective diagnostic tools, but more high-quality studies are needed for sufficient validation.
Data Availability
Not applicable.
Abbreviations
- AI:
-
Artificial intelligence
- NAFLD:
-
Non-alcoholic fatty liver disease
- NLR:
-
Negative likelihood ratio
- PLR:
-
Positive likelihood ratio
- DOR:
-
Diagnostic odds ratio
- CI:
-
Confidence interval
- NASH:
-
Non-alcoholic steatohepatitis
- MRI:
-
Magnetic resonance imaging
- US:
-
Ultrasound
- CT:
-
Computed tomography
- ML:
-
Machine learning
- DL:
-
Deep learning
- RF:
-
Radio frequency
- PDFF:
-
Proton density fat fraction
- PRISMA:
-
Preferred Reporting Items for Systematic Review and Meta-Analysis
- TP:
-
True positive
- TN:
-
True negative
- FP:
-
False positive
- FN:
-
False negative
- SROC:
-
Summary receiver operating characteristic curves
- AUC:
-
Area under the curve
References
Younossi ZM, Golabi P, Paik JM, Henry A, Van Dongen C, Henry L. The global epidemiology of nonalcoholic fatty Liver Disease (NAFLD) and nonalcoholic steatohepatitis (NASH): a systematic review. Hepatology (Baltimore MD). 2023;77(4):1335–47.
Friedman SL, Neuschwander-Tetri BA, Rinella M, Sanyal AJ. Mechanisms of NAFLD development and therapeutic strategies. Nat Med. 2018;24(7):908–22.
Wong VW, Chan WK, Chitturi S, et al. Asia-Pacific Working Party on Non-alcoholic Fatty Liver Disease guidelines 2017-Part 1: definition, risk factors and assessment. J Gastroenterol Hepatol. 2018;33(1):70–85.
Tapper EB, Lok AS. Use of Liver Imaging and Biopsy in Clinical Practice. N Engl J Med. 2017;377(8):756–68.
Regev A, Berho M, Jeffers LJ, et al. Sampling error and intraobserver variation in liver biopsy in patients with chronic HCV Infection. Am J Gastroenterol. 2002;97(10):2614–8.
Bedossa P, Carrat F. Liver biopsy: the best, not the gold standard. J Hepatol. 2009;50(1):1–3.
Piccinino F, Sagnelli E, Pasquale G, Giusti G. Complications following percutaneous liver biopsy. A multicentre retrospective study on 68,276 biopsies. J Hepatol. 1986;2(2):165–73.
Phisalprapa P, Supakankunti S, Charatcharoenwitthaya P, et al. Cost-effectiveness analysis of ultrasonography screening for nonalcoholic fatty Liver Disease in metabolic syndrome patients. Medicine. 2017;96(17):e6585.
Middleton MS, Van Natta ML, Heba ER, et al. Diagnostic accuracy of magnetic resonance imaging hepatic proton density fat fraction in pediatric nonalcoholic fatty Liver Disease. Hepatology (Baltimore MD). 2018;67(3):858–72.
Mollura DJ, Culp MP, Pollack E, et al. Artificial Intelligence in Low- and Middle-Income countries: Innovating Global Health Radiology. Radiology. 2020;297(3):513–20.
Zhang Y, Weng Y, Lund J. Applications of explainable Artificial Intelligence in diagnosis and Surgery. Diagnostics (Basel Switzerland) 2022; 12(2).
Han A, Byra M, Heba E, et al. Noninvasive diagnosis of nonalcoholic fatty Liver Disease and quantification of Liver Fat with Radiofrequency Ultrasound Data using one-dimensional convolutional neural networks. Radiology. 2020;295(2):342–50.
Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ (Clinical Research ed). 2021;372:n71.
Sounderajah V, Ashrafian H, Rose S, et al. A quality assessment tool for artificial intelligence-centered diagnostic test accuracy studies: QUADAS-AI. Nat Med. 2021;27(10):1663–5.
Whiting PF, Rutjes AW, Westwood ME, et al. QUADAS-2: a revised tool for the quality assessment of diagnostic accuracy studies. Ann Intern Med. 2011;155(8):529–36.
Yang B, Mallett S, Takwoingi Y, et al. QUADAS-C: a Tool for assessing risk of Bias in Comparative Diagnostic Accuracy studies. Ann Intern Med. 2021;174(11):1592–9.
Swets JA. Measuring the accuracy of diagnostic systems. Sci (New York NY). 1988;240(4857):1285–93.
Saba L, Dey N, Ashour AS, et al. Automated stratification of Liver Disease in ultrasound: an online accurate feature classification paradigm. Comput Methods Programs Biomed. 2016;130:118–34.
Minhas F, Sabih D, Hussain M. Automated classification of liver disorders using ultrasound images. J Med Syst. 2012;36(5):3163–72.
Li G, Luo Y, Deng W, Xu X, Liu A, Song E. Computer aided diagnosis of fatty liver ultrasonic images based on support vector machine. Annual International Conference of the IEEE Engineering in Medicine and Biology Society IEEE Engineering in Medicine and Biology Society Annual International Conference 2008; 2008: 4768-71.
Kuppili V, Biswas M, Sreekumar A, et al. Extreme Learning Machine Framework for risk stratification of fatty Liver Disease using Ultrasound tissue characterization. J Med Syst. 2017;41(10):152.
Hájek M, Dezortová M, Wagnerová D, et al. MR spectroscopy as a tool for in vivo determination of steatosis in liver transplant recipients. Magma (New York NY). 2011;24(5):297–304.
Byra M, Han A, Boehringer AS, et al. Liver Fat Assessment in Multiview Sonography using transfer learning with convolutional neural networks. J Ultrasound Medicine: Official J Am Inst Ultrasound Med. 2022;41(1):175–84.
Biswas M, Kuppili V, Edla DR, et al. Symtosis: a liver ultrasound tissue characterization and risk stratification in optimized deep learning paradigm. Comput Methods Programs Biomed. 2018;155:165–77.
Constantinescu EC, Udriștoiu AL, Udriștoiu ȘC, et al. Transfer learning with pre-trained deep convolutional neural networks for the automatic assessment of liver steatosis in ultrasound images. Med Ultrasonography. 2021;23(2):135–9.
Destrempes F, Gesnik M, Chayer B, et al. Quantitative ultrasound, elastography, and machine learning for assessment of steatosis, inflammation, and fibrosis in chronic Liver Disease. PLoS ONE. 2022;17(1):e0262291.
Owjimehr M, Danyali H, Helfroush MS. An improved method for Liver Diseases detection by ultrasound image analysis. J Med Signals Sens. 2015;5(1):21–9.
Sharma V, Juglan KC. Automated Classification of Fatty and Normal Liver Ultrasound Images Based on Mutual Information Feature Selection. IRBM 2018; 39(5): 313 – 23.
Ribeiro R, Tato Marinho R, Sanches JM. Global and local detection of liver steatosis from ultrasound. Annual International Conference of the IEEE Engineering in Medicine and Biology Society IEEE Engineering in Medicine and Biology Society Annual International Conference 2012; 2012: 6547-50.
Ribeiro RT, Marinho RT, Sanches JM. An ultrasound-based computer-aided diagnosis tool for steatosis detection. IEEE J Biomedical Health Inf. 2014;18(4):1397–403.
Acharya UR, Sree SV, Ribeiro R, et al. Data mining framework for fatty Liver Disease classification in ultrasound: a hybrid feature extraction paradigm. Med Phys. 2012;39(7):4255–64.
Zhou LQ, Wang JY, Yu SY, et al. Artificial intelligence in medical imaging of the liver. World J Gastroenterol. 2019;25(6):672–82.
Wei J, Jiang H, Gu D, et al. Radiomics in Liver Diseases: current progress and future opportunities. Liver International: Official Journal of the International Association for the Study of the Liver. 2020;40(9):2050–63.
Zhang L, Wang H, Li Q, Zhao MH, Zhan QM. Big data and medical research in China. BMJ (Clinical Research ed). 2018;360:j5910.
Spann A, Yasodhara A, Kang J, et al. Applying machine learning in Liver Disease and transplantation: a Comprehensive Review. Hepatology (Baltimore MD). 2020;71(3):1093–105.
Simon AB, Vitzthum LK, Mell LK. Challenge of directly comparing imaging-based diagnoses made by machine learning algorithms with those made by human clinicians. J Clin Oncology: Official J Am Soc Clin Oncol. 2020;38(16):1868–9.
Currie G, Hawk KE, Rohren E, Vial A, Klein R. Machine Learning and Deep Learning in Medical Imaging: Intelligent Imaging. J Med Imaging Radiation Sci. 2019;50(4):477–87.
Lee JG, Jun S, Cho YW, et al. Deep learning in Medical Imaging: General Overview. Korean J Radiol. 2017;18(4):570–84.
Shao L, Zhu F, Li X. Transfer learning for visual categorization: a survey. IEEE Trans Neural Networks Learn Syst. 2015;26(5):1019–34.
Ghassemi N, Shoeibi A, Khodatars M, et al. Automatic diagnosis of COVID-19 from CT images using CycleGAN and transfer learning. Appl Soft Comput. 2023;144:110511.
EASL Clinical Practice. Guidelines on non-invasive tests for evaluation of Liver Disease severity and prognosis – 2021 update. J Hepatol. 2021;75(3):659–89.
Nogami A, Yoneda M, Iwaki M, et al. Diagnostic comparison of vibration-controlled transient elastography and MRI techniques in overweight and obese patients with NAFLD. Sci Rep. 2022;12(1):21925.
Decharatanachart P, Chaiteerakij R, Tiyarattanachai T, Treeprasertsuk S. Application of artificial intelligence in non-alcoholic fatty Liver Disease and liver fibrosis: a systematic review and meta-analysis. Therapeutic Adv Gastroenterol. 2021;14:17562848211062807.
Jayakumar S, Sounderajah V, Normahani P, et al. Quality assessment standards in artificial intelligence diagnostic accuracy systematic reviews: a meta-research study. NPJ Digit Med. 2022;5(1):11.
Xu HL, Gong TT, Liu FH, et al. Artificial intelligence performance in image-based Ovarian cancer identification: a systematic review and meta-analysis. EClinicalMedicine. 2022;53:101662.
Wu MJ, Wang WQ, Zhang W, Li JH, Zhang XW. The diagnostic value of electrocardiogram-based machine learning in long QT syndrome: a systematic review and meta-analysis. Front Cardiovasc Med. 2023;10:1172451.
Korneev A, Lipina M, Lychagin A, et al. Systematic review of artificial intelligence tack in preventive orthopaedics: is the land coming soon? Int Orthop. 2023;47(2):393–403.
Funding
The present study is supported by the Natural Science Foundation of Shandong Province (grant no. ZR2021MH028), and the Medical Science and Technology Development Plan of Shandong Province (grant no. 202003031362).
Author information
Authors and Affiliations
Contributions
QZ and YDL designed the research. YDL, QZ and XJY acquired and analyzed the data. YDL and QZ wrote the manuscript. KW and YDL revised the final manuscript. All authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Ethics approval and consent to participate
This article did not involve human or animal subjects, and the ethics statement and consent to participates were not required.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Electronic supplementary material
Below is the link to the electronic supplementary material.
Supplementary Material 1: Table S1.
Search strategies; Table S2. Description of QUADAS-AI; Table S3. Performance of models from 15 included studies; Table S4. Pooled effects and Heterogeneity in subgroup analysis; Figure S1. Quality assessment of included articles using the QUADAS-AI criteria.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
About this article
Cite this article
Zhao, Q., Lan, Y., Yin, X. et al. Image-based AI diagnostic performance for fatty liver: a systematic review and meta-analysis. BMC Med Imaging 23, 208 (2023). https://doi.org/10.1186/s12880-023-01172-6
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s12880-023-01172-6