The liver is responsible for several functions in the body, including the primary detoxification of various metabolites, synthesis of proteins, and production of digestive enzymes (1); it also has a central role in carbohydrate and lipid metabolism. The primary functions of the liver include the production and excretion of bile and the detoxification and purification of the blood. Therefore, hepatic diseases need to be taken seriously. Liver diseases encompass a variety of diseases, especially metabolic dysfunction associated with fatty liver disease, alcohol-associated liver disease, viral hepatitis, and autoimmune liver diseases (2). These pathologies can cause fibrosis and may progress to cirrhosis, resulting in chronic liver disease (CLD). Cirrhosis is a major risk factor for the development of hepatocellular carcinoma (HCC) (3). Both are classified as global health burdens and together account for 3.5% of all deaths worldwide (4–6). In the clinic, a routine liver function of patients with CLD needs to be regularly controlled. Furthermore, for patients undergoing major liver resection, the function of the hepatic remnant needs to be assessed preoperatively to avoid posthepatectomy liver failure (PHLF). The latest advances in liver surgery and perioperative care have considerably improved patient outcomes following hepatectomy (7). Nevertheless, with morbidity rates of 10–40%, PHLF remains a major factor producing a poor prognosis (8). Therefore, precise preoperative assessment of liver function plays a crucial role in clinical decision-making. The need to measure liver function has led to the development of innovative methods to assess liver function. An outline of the newest methods is reviewed in this article. With a focus on the most promising MRI strategies, particular attention is given to the use of automation and artificial intelligence (AI) in liver imaging.

Liver biopsy is currently considered the gold standard for the evaluation of liver diseases. However, it has drawbacks, including invasiveness, costliness, and low patient acceptance. Furthermore, it is subject to sampling errors and examiner experience (9). In clinical practice, liver function and the severity of liver disease are evaluated based on clinical signs and biochemical blood parameters, such as bilirubin, alkaline phosphatase, glutamyl transferase, aspartate transaminase, alanine transaminase, albumin, and prothrombin time (1, 10). These are rather static tests reporting the current state of the liver without regarding changes in liver function (10). Nonetheless, in day-to-day routine testing, they are suitable for assessing the severity of liver damage and are thus an integral part of various scoring systems.

Comprehensive scoring systems for evaluating the liver function, such as the Child-Pugh (CP) score and the model for end-stage liver disease (MELD) score, have been developed. The CP Grading System is used for uniformly describing and classifying liver cirrhosis into different stages according to symptom severity. The ranking into the three CP groups (A-C) is based on a point scale (11); the CP score is calculated based on three objective [serum albumin, serum bilirubin, and international normalized ratio (INR)] and two subjective (ascites and encephalopathy) parameters. The subjective parameters vary with the use of diuretics or paracenteses in the treatment of ascites and the treatment of encephalopathy with lactulose (12). Therefore, in recent years, it has become common practice to use the MELD score to describe the severity of liver diseases (13), as no subjective parameters are considered. The MELD score is especially used in the allocation of organs for liver transplantation; it helps identify and prioritize the care of patients in acutely life-threatening situations due to liver disease and/or whose treatment is of utmost urgency. The MELD score is calculated using the following objective parameters: serum bilirubin, serum creatinine, and INR, from 6 to 40 points; the higher the score is, the lower the patient's probability of surviving the next 3 months without a liver transplant (14). The CP score and MELD score assess global liver function and are useful in determining whether patients with HCC and cirrhosis are candidates for resection or transplantation, but they are unable to determine the safe extent or removal (15). While they can roughly estimate the risks of performing a hepatectomy, they are not appropriate as a diagnostic tool in the preoperative environment.

To some extent, this also applies to the indocyanine green (ICG) test and the ¹³C-methacetin breath test (¹³C-MBT), dynamic tests that nevertheless are clinically useful in assessing global liver function. ICG clearance is currently the most widely used quantitative liver function test (16). ICG is a tricarbocyanine dye that binds to plasma proteins (albumin and α1-lipoprotein) and becomes evenly distributed in the blood within 2–3 min after intravenous injection. It is excreted into the bile exclusively via the liver without intrahepatic conjugation (17); its elimination is dependent on liver blood flow, hepatic cell function, and excretion via the biliary system. After administration, the blood ICG level decreases exponentially for ~20 min, at which time ~97% of the dye is excreted. ICG clearance is determined by serum sampling or pulse dye densitometry with an optical sensor on the finger (18, 19); commonly related parameters include the ICG retention ratio after 15 min (ICG-R15) and the plasma disappearance rate (ICG-PDR) (19). A large retrospective study showed that ICG clearance is associated with postoperative liver dysfunction; a PDR value <19.5% and an R15 value >5.6% were identified as cutoff values for identifying patients who are more likely to have a worse outcome for both minor and major hepatic resections (20). The ICG-R15 value corresponds to liver blood flow and hepatic functional reserve; in cirrhosis, it is used as a prognostic marker in decompensated cirrhotic patients and candidates for liver resection surgery (21). In hepatic surgery, such as liver resection and liver transplantation, the ICG elimination test is used as a liver function test to evaluate patient outcomes, as a prognostic marker, and as a diagnostic tool (22). However, the ICG test has limitations that hinder its use as a universal liver function test (23–26).

The ¹³C-MBT, like the ICG test, is a dynamic liver function test that reflects the patient's actual liver function at the time of the measurement. The ¹³C-MBT is based on the activity of the cytochrome P450 1A2 (CYP1A2) enzyme system, expressed exclusively and distributed evenly in the liver (27). The agent ¹³C-methacetin is metabolized exclusively by the CYP1A2 system, which converts it to paracetamol and ¹³CO₂. The exhaled ¹³CO₂ produces a change in the normal ¹³CO₂/¹²CO₂ ratio in the exhaled air and can be analyzed with an infrared spectroscopic detector. Therefore, the ¹³C-MBT provides quantitative information about liver function. The liver maximum capacity (LiMax) value (28) and a decision tree algorithm for hepatectomy that was developed from it (29) can be used to preoperatively evaluate a patient for liver surgery and better estimate the postoperative outcome. A preoperative LiMax value below 80 μg/kg/h for the future liver remnant increases the risk of PHLF (29). However, this is feasible only when the LiMax value is combined with, for example, CT volumetry to determine the volume of the liver and the future liver remnant. However, this test also has some limitations that need to be considered in its application (30). Regardless, a LiMax value above 315 μg/kg/h is assumed to indicate normal liver function, whereas a value below 140 μg/kg/h indicates a severe impairment of liver function (29). However, although the ¹³C-MBT, like the previously mentioned ICG test, provides functional data for the entire liver, it does not provide data for the functional activity of specific liver regions. The inhomogeneous distribution of liver function can be described by scintigraphic methods (31). For patients, a decisive disadvantage of scintigraphic imaging procedures is the constant risk of radiation exposure. In addition, several other liver function values can be found in the literature that has also found their way into clinical practice, such as the Makuuchi algorithm (decision algorithm for the surgical treatment of HCC) or values derived from FibroScan (ultrasound elastography) (32–35) (for an overview of the values, refer to Table 1). However, imaging techniques remain superior for visualizing the liver and provide additional important information about vascular anatomy. For instance, MRI techniques do not expose the patient to radiation and allow good visualization of organs and soft tissues. The development of liver-specific contrast agents such as gadoxetic acid has led to the development of several approaches for determining liver function that has been published in recent years. Moreover, studies have already shown that contrast-enhanced MRI is superior to CT (36) in the detection of intrahepatic recurrent HCC after surgery (37) and to dynamic CT in the detection and diagnosis of HCC (38).

Magnetic resonance imaging is capable of generating variable image contrast using different pulse sequences. In this process, the image parameters corresponding to the longitudinal (T1) and transverse (T2) relaxation times and the signal intensities on T1- and T2-weighted images vary depending on certain tissue properties (39). The intensity of these signals can be enhanced by contrast agents, such as gadolinium-based contrast agents, which have been established for use in liver MRI. Gadolinium (Gd) is a highly paramagnetic element that reduces the T1, T2, and T2^* relaxation times of surrounding water protons (40). Here, T1 shortening plays a very useful role in enhancing the signal intensity. In the clinical use for liver examination, there are two classes of contrast agents, namely, nonspecific extracellular and specific intracellular (hepatobiliary) agents. Nonspecific extracellular agents are taken up by the hepatic artery or portal vein, distribute rapidly in the extracellular space, and are almost exclusively excreted by glomerular filtration. They have no protein-binding properties and are used to assess the perfusion, blood flow, and vascularity of the liver (41). In contrast, hepatobiliary agents are taken up by functioning hepatocytes and excreted through the bile (41). This characteristic allows visualization of nonfunctioning hepatocyte lesions such as liver adenomas and HCC, which appear hypointense on hepatobiliary-phase images (Figure 1) and thus allow a functional assessment of the liver. Only two hepatobiliary agents, namely, gadoxetic acid (Gd-EOB-DTPA) and gadobenic acid (Gd-BOPTA), are in clinical use. However, Gd-BOPTA is less frequently used due to, among other reasons, its elimination half-life of 1–2 h and the fact that only 3–5% is cleared through biliary excretion. Gadoxetic acid enters hepatocytes via members of the organic anion transporting polypeptide (OATP) protein family; specifically, OATP1 B1 and B3 are responsible for transport into the liver (42). Biliary excretion from hepatocytes is performed by the multidrug resistance-associated protein 2 (MRP2) transporter. Approximately 50% of the administered Gd-EOB-DTPA is excreted via the kidney, and the rest is cleared via the OATP/MRP route (41). As a result of the greater hepatic uptake, liver parenchymal enhancement reaches its maximum (hepatobiliary phase, HBP) after 15–20 min, while for Gd-BOPTA, it starts after 1 h (40), making Gd-EOB-DTPA the preferred contrast agent. Consequently, both uptake and excretion of gadoxetic acid allow the quantification of regional liver function. However, it should be mentioned that both uptake and excretion can be influenced by the altered expression of OATPs and MRPs, either due to genetic factors or liver disease (43–47). Nonetheless, in early studies with Gd-EOB-DTPA, it became clear that enhancement could be useful for detecting liver lesions in HBP images (48). In 2010, Tajima et al. first suggested that the degree of enhancement may reflect liver cellular function (49).

Contrast-enhanced MRI enables both the characterization of liver lesions and the assessment of regional liver function. In addition to functional information, it provides important anatomical information, e.g., for surgical planning, including lesion volume and vascular supply. In other words, contrast-enhanced MRI could be used as a one-stop examination to assess liver function. Recent relevant literature reveals a variety of equations and names for liver function measurements based on contrast-enhanced MRI data, which can be somewhat confusing at first glance. Essentially, however, two parameters are measured for assessing liver function, namely, signal intensities or the T1 relaxation time.

Several steps that need to be developed in terms of automating image analysis to quantify liver function are as follows: (A) liver segmentation and volumetry; (B) image registration of native and contrast-enhanced scans; (C) detection, scoring, and quantification; and (D) fully automated MRI image-based liver function quantification (Figure 2). Both conventional image processing methods and novel AI approaches are suitable for this purpose. AI applications are increasing in popularity in medical research; further insight into AI in medical imaging is provided in the Excursus Box.

When searching PubMed with the terms MRI, liver, and CNN, approximately 50 articles published in the last few years were retrieved, highlighting the novelty of the field. These studies reveal that the dominant topics are segmentation and classification for diagnostic support. For instance, both liver segmentation and volume determination can be performed automatically, with an intraclass correlation coefficient of 0.987 (86). Likewise, acceptable Dice coefficients comparing AI with manual segmentation between 0.91 and 0.95 have been observed (87–89). Even approaches using the same liver segmentation model for different imaging modalities, such as CT and MRI, have been successful (90). Other studies have focused more on the detection or classification of liver lesions or disease, which is important for the development of computer-aided detection or diagnosis (CADe and CADx) systems. By using CNN models, one study showed that it was possible to distinguish fibrosis stages F2 to F4 with high diagnostic performance (AUC: F4, 0.84; F3, 0.84; and F2, 0.85) (91). CNN models can also detect liver cirrhosis at the expert level, indicating the feasibility of assisting the radiologist in diagnosis (92). Beyond that, however, the majority of studies focus on the differentiation of liver lesions (87, 93, 94) or detection (94, 95) or grading of HCC (96). These proof-of-concept studies describe good-performing models with AUCs over 0.90 in some cases and thus the potential to assist radiologists in diagnosis. Although none of these studies address the determination of liver function, they demonstrate the value of AI approaches in liver diagnostics using radiological methods. Consequently, it is assumed that future research and publications will address the automation of MRI image-based liver function quantification.

One point that has not yet been addressed is the duration and cost-effectiveness of MRI procedures. The sequences needed for T1 relaxometry lengthen the patient's examination by <1 min. However, it may take time even for an experienced radiologist to manually determine SI or T1 relaxometry. Given the AI-based IT solutions outlined earlier, it can be assumed that this will one day be an automatic process. In addition to personnel costs, there are also licensing costs to carry out T1 relaxometry. The T1 relaxometry sequences can be easily incorporated into the existing liver MR exam as a part of the patient care, eliminating the need for an additional exam. However, although the aforementioned examples of AI in liver diagnosis and imaging illustrate the power of these approaches, they have not yet found their way into clinical practice, in part due to the lack of clinical validation. In addition, open-source data are desirable to allow objective evaluation and comparison of different methods and approaches, which is not yet possible. Thus, the answer to our initial question, “where do we stand in determing liver function?,” is as follows: liver function determination by MRI is possible, and rrT1 seems to be the best candidate. Although the presented studies demonstrate that values (SI, RE, and rrT1) obtained from contrast-enhanced MRI are significantly correlated with global liver function parameters, there is no cutoff value for stratifying patients thus far. Therefore, large prospective studies are needed to establish them as functional parameters, as most works to date have been based on retrospective studies. Of course, it would be desirable to implement software capable of segmentation and determining diagnostic values such as rrT1. However, given the rapid development of AI software in recent years, this is likely to be a minor problem moving forward.

CR and PW conceived this manuscript. All authors contributed to the content and co-authored the manuscript. Each author agrees to be responsible for the content work. All authors contributed to the article and approved the submitted version.

The authors are part of a collaborative project sponsored by the German Federal Ministry of Education and Research as part of the program “Image-based diagnostics” (Grant number: 13GW0363A-C).

MS, JP, NK, RB, and JW were employed by MeVis Medical Solutions AG. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.