ACHAIKI IATRIKI | 2025; 44(1):12–15
Editorial
Efstratios Syrmas1, Ilias Gatos1, Paraskevi F. Katsakiori2, Stavros Tsantis1, Stavros Spiliopoulos3, George C. Kagadis1
13DMI Research Group, Department of Medical Physics, School of Medicine, University of Patras, Rion, Greece
2Health Center of Akrata, Akrata, Greece
3Second Department of Radiology, School of Medicine, University of Athens, Athens, Greece
Received: 28 Mar 2024; Accepted: 05 Apr 2024
Corresponding author: George C. Kagadis, PhD, FAAPM, Professor of Medical Physics – Medical Informatics, Department of Medical Physics, School of Medicine, University of Patras, Rion, GR 26504, Greece, Tel.: +30 2610 962345, e-mail: gkagad@gmail.com
Key words: Chronic liver disease, liver ultrasound elastography, artificial intelligence, machine learning, deep learning, diagnostic accuracy
INTRODUCTION
Chronic Liver Disease (CLD) is a leading public health concern [1]. CLD progresses through inflammation to fibrosis, and – if left untreated – to cirrhosis. Cirrhosis, the end-stage of the disease, may lead to hepatocellular carcinoma, liver failure, portal hypertension and eventually death. Accurate diagnosis of CLD is essential to secure effective clinical management and intervention strategies. Liver biopsy (LB) is considered the ‘Gold Standard’ for CLD diagnosis as it provides direct and detailed histological information. Nonetheless, LB is invasive and prone to sampling errors leading to significant inter- and intra-observer variability. These limitations have stimulated the quest for less or non-invasive diagnostic approaches leading to the adoption of elastography that demonstrates high correlation between liver stiffness and liver fibrosis [2]. Several literature reports aim to demonstrate the accuracy of this correlation. Their outcome is to calculate the liver stiffness cut-off values with the aid of ROC analyses. Fibrosis stages should be differentiated optimally to offer the radiologist a simple tool that corresponds stiffness values to fibrosis stages using a certain examination protocol [3].
Two modalities are mainly employing elastography, MR-Elastography (MRE) and Ultrasound Elastography (USE) [4]. Both have gained popularity due to their easy applicability and high-performance in differentiating various severity stages of CLD. Elastography constitutes a rather recent non-invasive modality for the assessment of liver fibrosis and has already started revolutionizing CLD diagnosis. Elastography basic principle is to generate a vibration within the tissue of interest, record the vibration’s propagation through the tissue and subsequently deduce elasticity from the tissue response. Ultrasound (US) system manufacturers have used different technological interpretations of this principle leading to US systems that are different in terms of use in clinical practice.
While there is mainly one MRE variant, multiple USE variants are nowadays commercially available including Vibration Controlled Transient Elastography (VCTE), widely known as Fibroscan, Real Time Elastography (RTE), Acoustic Radiation Force Impulse (ARFI) Elastography, Shear Wave Elastography (SWE) and Sound Touch Elastography (STE) to name a few. All these techniques (except for RTE which makes a qualitative relative elasticity estimation) make a quantitative tissue stiffness estimation in an area of interest. Currently, most of these variants provide a colored elasticity map to visually guide the examiner to an optimum measurement. These techniques have been extensively studied and demonstrate USE’s high diagnostic performance in CLD fibrosis stage differentiation.
USE techniques show certain limitations such as significant inter- and intra-observer variability, best stiffness cut-off values overlap between studies, and presence of non-liver fibrosis related factors that affect stiffness measurements leading to over- or under-estimation of patient’s clinical condition. To overcome these limitations, Artificial Intelligence (AI) has recently been used and boosted the performance of computer aided diagnosis (CAD) systems in the pursuit of accurate CLD stage assessment. AI algorithms achieve – or even outperform – experts’ accuracy in CLD assessment with USE, rendering them a useful tool in clinical practice [5]. In this editorial, the current state of AI applications along with their challenges and future perspectives in USE for CLD assessment are presented.
Main Body
AI applications can be categorized in Machine Learning (ML) and Deep Learning (DL) based ones [6]. ML requires feature extraction (radiomics in the case of radiological features) and manipulation from raw data to model input. DL directly evaluates raw data bypassing manual or semi-automated feature extraction and analysis for model input. In the case of CLD assessment various studies exist that deploy ML or DL models and attempt to address USE limitations or further improve diagnostic accuracy.
Machine Learning Studies
Few approaches with the use of image processing and analysis for feature extraction from US images have been proposed in ML studies. Gatos et al. made an inverse Red-Green-Blue (RGB) to stiffness mapping of 2D elastogram of SWE images and extracted and analyzed features from the resulting region of interest (ROI). They fed a Support Vector Machine (SVM) with the extracted features and differentiated CLD patients from healthy subjects with high accuracy surpassing clinical studies’ performance [7, 8]. Furthermore, they suggested specific feature combinations and value ranges to indicate fibrosis existence [8]. More analyses on other fibrosis stage groups and their differentiation are deemed necessary to complete the deployed algorithms’ potential on fully assessing CLD fibrosis staging.
Durot et al. also noted that SVMs, a multimodel ML algorithm, can effectively grade liver fibrosis through USE [9]. This approach showed diagnostic performance comparable to MRE, further broadening the scope of non-invasive liver fibrosis assessment tools. A hybrid ML methodology, combining a Convolutional Neural Network (CNN) with dual classifiers – SoftMax and SVM – was proposed by Sattar Jabbar et al. for the identification of liver fibrosis through the analysis of 700 US shear wave elastography images [10].
Deep Learning Studies
Several studies have recently shown that DL is a powerful tool with increased diagnostic accuracy over clinical or ML studies. Gatos et al. employed an elastogram reliability tool that temporally excluded unstable areas of the image [11]. Afterwards, they compared the examiners and DL performance on both filtered and full elastograms. Their results indicated that the examiners’ measurements’ accuracy was poor on the excluded areas. However, on the areas left intact, performance was relatively accurate. DL showed marginal improvement on performance when fed with the filtered images. Kagadis et al. further explored the diagnostic performance on the filtered and non-filtered images on a variety of settings and DL schemes validating their superior performance over clinical and ML approaches [12].
Subsequently, the implementation of DL radiomics of elastography demonstrated superior accuracy compared to traditional methods for accurately staging liver fibrosis in chronic hepatitis B patients through non-invasive 2D-SWE image analysis as shown by Wang et al. [13]. Xue et al. employed transfer learning to analyze elastogram and grayscale US images that further improved diagnostic accuracy, demonstrating the benefit of integrating both modalities over using them separately [14]. Meng et al. developed a liver fibrosis classification method using transfer learning with VGGNet and a deep classifier, FCNet, for ultrasound elastography images [15].
Challenges
Application of AI tools in medical imaging and diagnosis has been accelerated and facilitated the emergence of new pathways of optimizing CLD prognosis and management [16]. However, a few challenges need to be considered before such AI tools become fully operational in the clinical set-up. Although the integration of AI into hepatic elastography seems promising, it encounters several limitations. The robustness of AI models is often reduced when faced with data from diverse patient populations or imaging systems. These include data heterogeneity and quality issues arising from varied acquisition protocols and operator techniques, which challenge AI’s ability to generalize. The ‘black box’ nature of DL models complicates their interpretability, a critical factor for clinical acceptance. Furthermore, the absence of standardized validation protocols for USE makes benchmarking AI tools challenging [17]. USE is used in real-time clinical procedures, requiring immediate analysis and interpretation. Integrating AI to enhance or automate this process demands high computational efficiency to provide instant feedback without disrupting the clinical workflow. Additionally, when introducing AI, it is important to deal with complicated rules and ethical issues to keep patients safe and their information private [18]. We also need to conclude on common rules for testing these AI systems in USE.
Future Directions
A complex and careful approach is deemed necessary to overcome the limitations faced by AI in hepatic elastography. Firstly, enhanced representability of the sample used can be achieved through inter-institutional collaboration to create large, diverse, and well-annotated datasets. This advancement could contribute to the development of AI models that exhibit improved generalizability and robustness, particularly in varying USE techniques and patient populations, ensuring consistent performance across different clinical settings. Secondly, enhancing AI model interpretability can involve incorporating explainable AI (XAI) techniques in the case of USE. These methods may explain the AI’s decision-making pathways in elastographic analysis and offer a clearer understanding of its diagnostic predictions. Thirdly, a collaborative effort is needed to establish standardized validation protocols. These standards will help make sure that AI tools are reliable and useful in medical practices. Finally, general dealing with the detailed rules and ethical issues requires a firm commitment to privacy standards like the General Data Protection Regulation (GDPR), and a focus on ethical guidelines that prioritize patient safety and data security.
To conclude, current data demonstrate that AI could be implemented to improve the diagnostic accuracy of USE and avoid a great number of liver biopsies, which are related to low but not insignificant morbidity and mortality, and their use should be carefully reconsidered especially in more sensitive subgroups such as the pediatric population [19, 20]. Further investigation is required to validate these initial results and address current issues as to introduce AI-assisted hepatic elastography into everyday clinical practice.
Conflict of Interest: There is no conflict of interest
Declaration of Funding Sources: This study was financed by The Hellenic Foundation for Research and Innovation (H.F.R.I.) under the ‘2nd Call for H.F.R.I. Research Projects to support Faculty Members & Researchers’ (Project Number: 2692).
Author Contributions: EF, IG, GCK conceived idea; EF, IG, PFK, ST, SS drafted manuscript; GCK critically copy-edited manuscript; EF, PFK, GCK revised manuscript; GCK oversaw study.
REFERENCES
- Marcellin P, Kutala BK. Liver diseases: A major, neglected global public health problem requiring urgent actions and large-scale screening. Liver Int. 2018;38(Suppl 1):2-6.
- Sumida Y, Nakajima A, Itoh Y. Limitations of liver biopsy and non-invasive diagnostic tests for the diagnosis of nonalcoholic fatty liver disease/nonalcoholic steatohepatitis. World J Gastroenterol. 2014;20(2):475-85.
- Mueller S, Sandrin L. Liver stiffness: a novel parameter for the diagnosis of liver disease. Hepat Med. 2010;2:49-67.
- Yin M, Venkatesh SK. Ultrasound or MR elastography of liver: which one shall I use? Abdom Radiol (NY). 2018;43(7):1546-51.
- Zhou LQ, Wang JY, Yu SY, Wu GG, Wei Q, Deng YB, et al. Artificial intelligence in medical imaging of the liver. World J Gastroenterol. 2019;25(6):672-82.
- Sarker IH. AI-Based Modeling: Techniques, Applications and Research Issues Towards Automation, Intelligent and Smart Systems. SN Comput Sci. 2022;3(2):158.
- Gatos I, Tsantis S, Spiliopoulos S, Karnabatidis D, Theotokas I, Zoumpoulis P, et al. A new computer aided diagnosis system for evaluation of chronic liver disease with ultrasound shear wave elastography imaging. Med Phys. 2016;43(3):1428-36.
- Gatos I, Tsantis S, Spiliopoulos S, Karnabatidis D, Theotokas I, Zoumpoulis P, et al. A Machine-Learning Algorithm Toward Color Analysis for Chronic Liver Disease Classification, Employing Ultrasound Shear Wave Elastography. Ultrasound Med Biol. 2017;43(9):1797-810.
- Durot I, Akhbardeh A, Sagreiya H, Loening AM, Rubin DL. A New Multimodel Machine Learning Framework to Improve Hepatic Fibrosis Grading Using Ultrasound Elastography Systems from Different Vendors. Ultrasound Med Biol. 2020;46(1):26-33.
- Sattar Jabbar Z, Qusai Al-Neami, A., Khawwam, A.A., and Munther Salih, S. Liver fibrosis processing, multiclassification, and diagnosis based on hybrid machine learning approaches. Indones J Electr Eng Comput Sci. 2023;29(3):1614.
- Gatos I, Tsantis S, Spiliopoulos S, Karnabatidis D, Theotokas I, Zoumpoulis P, et al. Temporal stability assessment in shear wave elasticity images validated by deep learning neural network for chronic liver disease fibrosis stage assessment. Med Phys. 2019;46(5):2298-309.
- Kagadis GC, Drazinos P, Gatos I, Tsantis S, Papadimitroulas P, Spiliopoulos S, et al. Deep learning networks on chronic liver disease assessment with fine-tuning of shear wave elastography image sequences. Phys Med Biol. 2020;65(21):215027.
- Wang K, Lu X, Zhou H, Gao Y, Zheng J, Tong M, et al. Deep learning Radiomics of shear wave elastography significantly improved diagnostic performance for assessing liver fibrosis in chronic hepatitis B: a prospective multicentre study. Gut. 2019;68(4):729-41.
- Xue LY, Jiang ZY, Fu TT, Wang QM, Zhu YL, Dai M, et al. Transfer learning radiomics based on multimodal ultrasound imaging for staging liver fibrosis. Eur Radiol. 2020;30(5):2973-83.
- Meng D, Zhang, L., Cao, G., Cao, W., Zhang, G., and Hu, B. Liver Fibrosis Classification Based on Transfer Learning and FCNet for Ultrasound Images. IEEE Access. 2017;5:5804-10.
- Lee HW, Sung JJY, Ahn SH. Artificial intelligence in liver disease. J Gastroenterol Hepatol. 2021;36(3):539-42.
- Myllyaho L, Raatikainen, M., Mannisto, T., Mikkonen, T., and Nurminen, J.K. Systematic literature review of validation methods for AI systems. J Syst Softw. 2021;181:111050.
- Farhud DD, Zokaei S. Ethical Issues of Artificial Intelligence in Medicine and Healthcare. Iran J Public Health. 2021;50(11):i-v.
- Neuberger J, Patel J, Caldwell H, Davies S, Hebditch V, Hollywood C, et al. Guidelines on the use of liver biopsy in clinical practice from the British Society of Gastroenterology, the Royal College of Radiologists and the Royal College of Pathology. Gut. 2020;69(8):1382-403.
- Ovchinsky N, Moreira RK, Lefkowitch JH, Lavine JE. Liver biopsy in modern clinical practice: a pediatric point-of-view. Adv Anat Pathol. 2012;19(4):250-62.