Lessons learnt from pathologic imaging correlation in the liver: an historical perspective

Abstract

Imaging and pathology can be considered as two sides of the same diagnostic coin. Yet, pathology remains the gold-standard technique for the diagnosis of most diseases. Nevertheless, significant and constant progress in imaging has been made thanks to fruitful rad-path correlations. The aim of this article is to show how much imaging has benefited from pathology and to illustrate the different ways in which imaging has evolved according to different types of pathological references. Imaging of hepatocellular carcinoma shows how image-based knowledge and expertise can be exploited to yield a non-invasive diagnosis approaching that of a fixed, robust pathological reference. Hepatocellular adenomas provide an example of the constant radiological evolutions triggered by changing pathological definitions. Finally, hepatic steatosis illustrates the possibility for imaging to surpass its historical reference, and become a new gold-standard. For these three examples, we have taken a historical approach to demonstrate how rad-path interminglement creates knowledge.

INTRODUCTION

Imaging and pathology have become central to the management of patients, from diagnosis to treatment. These two disciplines are complementary in providing multiscale and multiparametric characterization of tissues and organs of interest. Image interpretation is at the centre of both and relies on medical expertise, so much so that both can be considered as two sides of the same diagnostic coin. Yet, they don’t have the same place in patient management: while imaging is now of utmost importance in the diagnostic process, pathology is almost always needed when a final diagnosis is to be reached. Consequently, two things are deeply engraved in the mind of radiologists, as it is in that of clinicians: first, that pathology has something to do with the ground truth, and second, that radiologists consider it as a reference method to which they must compare their performance. Pathology has to do with the ontology of medicine (what is and what is not), and with its epistemology (what’s true and what’s not). Modern medicine is not the daughter of the anatomoclinical method of Laennec, Charcot, and Bichat for nothing.

Of course, truth comes at a certain cost, and pathology is based—so far—on invasive procedures, from fine needle aspiration to biopsy, resection, and sometimes autopsy. Aside from its inherent frequent retrospectiveness, the associated harm that this invasiveness necessarily entails is, by far, compensated for by the benefit the medical community gains from it. It also constitutes a major stimulation for academic imaging and a bridge to molecular biology. All innovations in pathology (description of new entities, subdivisions, refinement of classifications, etc.) rapidly translate into major radiological improvements. Imaging has come a long way since the first morphological descriptions of macroscopic alterations at the organ-level. It is now capable of detecting and depicting molecular alterations. Of course, technological improvements in the hardware and software we use have been tremendous, enabling us as radiologists to provide answers that are immensely more detailed and refined. Meanwhile, the depth of pathological understanding of diseases (from morphological pathology, to immunohistochemistry, or molecular biology) has increased exponentially too. Not only are our answers more pertinent but also the very questions posed are far more precise and relevant.

Considering pathology as ground truth has a downside: it is frequently seen as more rigid and straightforward than it really is. Nicolas Boileau, a French poet, said, “whatever is well conceived is clearly said, and the words to say it flow with ease”. We inherited this idea that truth has something to do with obviousness. This obviates how complex, changeable, and uncertain pathology may sometimes be. It is a field of knowledge and expertise submitted, as any other, to dynamic collective movements, inner struggles, some very deep and some just scratching the surface. For radiologists it means that pathology is not to be worshiped. Of course it can make mistakes, but much more importantly it can eventually be a suboptimal reference for us. Like the student surpassing their teacher, it may often be that imaging surpasses pathology.

The aim of this article is to illustrate different types of rad-path relationships, and to show how much imaging has benefited from pathology, using several examples taken from the field of hepatology.

From reference to complement: the concept of non-invasive diagnosis

The vast majority of liver diseases require pathological assessment: patients with chronic liver disease undergo biopsy to quantify the amount of fibrosis and assess the presence of fat or inflammation, and focal liver lesions require pathological confirmation. Yet, in daily practice, an invasive diagnosis is not justified in all cases for patient management. Indeed, a significant proportion of patients can be diagnosed using imaging only, based on the key-concept of “non-invasive diagnosis”. This concept is probably well-known by most readers, but we need to examine its meaning, together with its inherent conditions of possibility, and the consequences for patients.

The concept of non-invasive diagnosis

“Non-invasive diagnosis” refers to the possibility of reaching a definitive diagnosis without the need for invasive procedures. Translated into the liver rad-path context, it means that imaging can diagnose liver diseases or hepatic tumors without the need for liver biopsy. We can summarize the frame of this approach in three points (Figure 1):

Figure 1.

Schematic representation of the different steps of imaging-based non-invasive diagnosis. Applied in a specific clinical context (1), a combination of imaging features (2) allows for the differentiation between patients with and without a certain diagnosis (3). When the diagnosis is certain, patients can be managed without pathological confirmation (4). When it is not, further investigations are required that may (or may not) include pathological assessment.

• It implies that combinations of imaging features that are associated with a given disease are considered specific enough (ideally pathognomonic) for the diagnosis.

• A fundamental pre-requisite is that such combinations can be applied in clearly identified clinical contexts only.

• The consequence is that patients are classified into groups following a rule-in/intermediate/rule-out approach that will modulate the need for further invasive investigations.

A good example is the guidelines for the non-invasive diagnosis of hepatocellular carcinoma (HCC), introduced and updated on a frequent basis by the European Association for the Study of the Liver (EASL).

The example of hepatocellular carcinoma

As a general oncologic rule, the diagnosis of cancer needs to be certain, given the consequences for the patient in terms of treatment-related toxicity and complications, which are only acceptable if counterbalanced by significant benefits in terms of survival. Therefore, any suspected tumor should undergo tissue sampling and pathological assessment. Yet, according to the 2018 EASL guidelines (and in line with the 2012 EASL guidelines), in the setting of cirrhosis, the diagnosis of HCC can be reached non-invasively by imaging if a lesion >10 mm in size shows typical features, namely a combination of hypervascularity in the late arterial phase [defined as arterial phase hyperenhancement (APHE) according to LI-RADS (Liver Imaging Reporting and Data System)] classification¹ and washout on portal venous and/or delayed phases on CT or MRI using extracellular contrast agents. This corresponds to changes in hepatic microcirculation, which occur during the development of HCC.^2,3 Despite minor differences, this has been also endorsed by other hepatology societies from both the West and the East.^4–7

This statement is so widely accepted, and may seem so obvious, that one may forget that it is the result of a long, complex collaborative effort based on rad-path correlations.

Historically, pathologists have shown that HCC are subject to significant neo-angiogenesis. Its occurrence is considered to be the turning point in liver hepatocarcinogenesis. It is visible as contrast uptake on contrast-enhanced imaging during the arterial phase. This is why in 2001, experts issued the first set of non-invasive criteria stating that the diagnosis of HCC could be made if the presence of arterial phase hyperenhancement was present on two different imaging modalities (CT, MRI, angiography or ultrasound) in cirrhotic patients with nodules >2 cm.⁸ The very high pre-test probability of HCC was the reason why such non-invasive criteria were restricted to cirrhotic patients. The main problem was that numerous other lesions and pseudolesions may cause focal APHE in cirrhotic patients. For instance, in a retrospective study published in 2003 that included patients undergoing CT before liver transplantation, Freeny et al described 61 nodules showing APHE, among which only 17 (28%) were HCCs.⁹ Other nodules were mostly benign regenerative or dysplastic nodules.⁹ This criterion (APHE) clearly lacked specificity.

As a consequence, in 2005 experts from both EASL and the American Hepatology society (AASLD) modified the previous guidelines by including a new HCC radiological hallmark, i.e. washout on the portal venous and/or delayed phase in order to gain specificity. Non-invasive diagnosis of HCC was therefore established in nodules > 20 mm showing both APHE and washout.¹⁰ For nodules ranging from 10 to 20 mm, hallmarks had to be depicted on two imaging modalities [CT, MRI, or contrast-enhanced ultrasound (CEUS)] (Figure 2). Things could have been considered as definitive but numerous rad-path correlation studies have since challenged these criteria, leading to substantial updates and modifications regarding three points: (1) performance of imaging in nodules measuring 10–20 mm, (2) comparison of different imaging modalities, and (3) role of liver specific MR contrast agents.

Figure 2.

A 64-year-old male patient with HCV-related cirrhosis. During ultrasound surveillance, a focal lesion was depicted in the hepatic dome, and the patient underwent contrast-enhanced CT for characterization. The lesion shows both APHE (arrow in A) and washout on delayed phase images (arrow in B), which allows for the non-invasive diagnosis of HCC. Another lesion showing APHE is depicted in the left lobe (arrow in C). This lesion does not display washout on portal (not shown) or delayed phase images (arrow in D), and corresponded to an arterioportal shunt. Pathological analysis with hematoxylin eosin staining (E) showed a well-differentiated form of HCC with trabecular pattern. Tumoral hepatocytes displayed enlarged nuclei with prominent nucleoli. CD31 immunostaining (F) highlights the peritumoral endothelial lining. APHE, arterial phase hyper enhancement; HCC, hepato cellular carcinoma.

Regarding the challenging group of 10–20 mm nodules, a study addressing the diagnostic accuracy of MRI in a series of transplanted patients reported a false-positive rate of more than 10% when using only one imaging technique.¹¹ These results supported the use of more than one imaging technique, but prospective studies have shown that combining two imaging modalities warranted high specificity at the cost of very low sensitivity (around 30%). The consequence was that more than two-thirds of nodules would still require pathological exploration.¹² Sersté et al published a prospective study reporting a false-positive rate above 10% with either one or two imaging techniques, with a specificity of 81 and 85%, respectively.¹³ This explains why in 2012, EASL guideline experts recommended a mixed approach for 10–20 mm nodules, advocating the use of two coincidental techniques in non-expert centers, and only one in expert centers.² However, this statement was not evaluated and expert centers were not defined.

In 2010, Sangiovani et al published a study suggesting that the use of a sequential algorithm would maintain an excellent specificity but increase the sensitivity, with significant savings in terms of liver biopsy for 10–20 mm nodules.¹⁴ This elegant approach also takes into consideration the respective value of different imaging techniques (MRI, CT or CEUS). This was confirmed by a recent prospective multicentric study, including 381 patients with 544 nodules.¹⁵ This study reported a sensitivity and specificity of 70.6 and 83.2% for MRI using extracellular contrast agents, and 67.9 and 76.8% for CT, respectively in 10–20 mm nodules.¹⁵ Here again, the coincidental use of CT and MRI had a specificity of 100%, but a sensitivity of 55.1% only. Yet authors advocated for sequential use of imaging modalities. Interestingly, if CEUS as a diagnostic tool had initially been questioned because of the risk of false-positives, authors of the above-mentioned study showed that a very high specificity was reached when using CEUS as a second-line examination after a first inconclusive CT or MRI.¹⁵ This became possible only because other studies had refined and identified the typical hallmarks for HCC at CEUS (i.e. APHE and late (>60 s) and mild washout.^16,17 A recent large retrospective study that included more than 1000 lesions in patients with cirrhosis showed that this HCC CEUS pattern had close to 100% positive-predictive value.¹⁸ This approach has been endorsed by EASL in the most recent 2018 version of their guidelines.³

Another very important development was the introduction of liver specific MR contrast agents. These agents are gadolinium chelates that are taken up by functioning hepatocytes. Rad-path studies have shown that their internalization is mediated by organic anionic transporting polypeptides (OATP) expressed on the sinusoidal membrane of functional hepatocytes.¹⁹ The level of expression of these proteins is significantly decreased in impaired hepatocytes. As a consequence, these contrast agents have been shown to be accurate markers of hepatocellular function. Their injection allows for the acquisition of dedicated T ₁ weighted images obtained when the liver and bile ducts are markedly enhanced [the so-called “hepatobiliary phase” (HBP)]. On HBP images, non-hepatocellular tumors, tumors containing impaired hepatocytes (such as HCC), but also vessels and cysts appear hypointense. Importantly, loss of hepatocellular function occurs early during the carcinogenesis of liver tumors, even before tumor neoangiogenesis. This is why 80 to 90% of HCC are hypointense on HBP, while most non-HCC, cirrhosis-associated regenerative or dysplastic nodules appear iso- or hyperintense. Therefore, it has been shown that liver specific MR agent-enhanced MRI has a higher sensitivity for detecting nodules. A recent meta-analysis focusing on the diagnostic performance of MRI for diagnosing HCC up to 2 cm has shown that gadoxetic acid enhanced-MRI had significantly increased sensitivity compared to extracellular contrast agent MRI (92 and 67%, respectively).²⁰ It may be tempting to consider this new feature (hypointensity on HBP) as a new imaging hallmark of HCC, and to expect improvement of diagnostic performance, the same way introducing the criterion of washout did. This approach was recently adopted by Renzulli et al in a study including 420 nodules >1 cm in 228 patients.²¹ Authors built a classification and regression tree using three MRI findings that were independently associated with HCC (i.e. HBP hypointensity, arterial hyperintensity, and diffusion restriction). This algorithm demonstrated, both in the entire study population and for nodules ≤2 cm, higher sensitivity but slightly lower specificity than classical criteria.²¹ Yet, the significance of hypointensity on the transitional phase and/or the HBP is not the same as that of washout. This why other studies have shown that hypointensity on HBP used as an alternative to washout led to a significant increase in sensitivity for the diagnosis of HCC, but at the cost of an unacceptable decrease in specificity.^22,23 The 2018 EASL guidelines authorize the use of liver-specific MR agents, but restrict the definition of washout to the portal venous phase only.⁴

This journey through the history of the non-invasive diagnosis of HCC illustrates the first type of rad-path relationship: starting with a fixed and consistent pathological reference, imaging works at defining the narrowest possible non-invasive definition of a disease, and to apply it to the widest possible patient population, sometimes replacing pathology, and, when not possible, acting as its complement. The resulting compromise is challenged by further refinement of rad-path correlations and technical developments, in the form of a changing collective knowledge.

Advances in pathology stimulate radiological progress

The previous developments started with a clearly defined pathological reference, and we tried to show how imaging could progressively improve by confronting it. Another completely different situation occurs when the pathological reference keeps changing. Here, radiological developments are constantly challenged and are required to keep up with the pathological pace. Imaging of hepatocellular adenomas offers a good example.

The example of hepatocellular adenoma

At the beginning of the 2000s, radiologists would learn that hepatocellular adenomas (HCAs) are rare benign tumors, occurring more frequently in young adult females taking oral contraception. They would also learn that these tumors were to be treated because they exposed patients to complications such as bleeding or malignant transformation. From a radiological point of view, they would know that HCA are heterogeneous lesions, showing various degrees of APHE and washout, frequently containing fat and possibly intratumoural hemorrhage. Overall, one tumor with several presentations. At this time, they also knew that HCA were to be distinguished from focal nodular hyperplasia (FNH), for which non-invasive diagnosis was possible in defined contexts. At this time, telangiectatic FNH was described and was considered as a variant of FNH.

In the mid-2000s, two important articles were published that completely changed things. First, Paradis et al demonstrated in 2004 that telangiectatic FNH display a molecular pattern closer to that of HCA than to FNH, and suggested that these lesions should instead be referred to as “telangiectatic hepatocellular adenomas”.²⁴ Second, Zucmann-Rossi et al published a seminal study in 2006 demonstrating that genotype-phenotype correlations could identify several HCA subtypes²⁵:

• HCA inactivated for HNF-1a (H-HCA), accounting for 30 to 40% of HCA, and characterized by marked steatosis, lack of cytological abnormalities, and no inflammatory infiltrates.

• Inflammatory HCA (I-HCA), accounting for 40–55% of HCA, corresponding to the previous telangiectatic FNH, with frequent cytological abnormalities, dystrophic vessels, and inflammatory infiltrates associated with foci of sinusoidal dilatation.

• Beta-catenin activation HCA (b-HCA), accounting for 10–20% of HCA, with frequent cytological abnormalities and pseudoglandular formation. These lesions have a higher risk of malignant transformation.

• Unclassified HCA, accounting for 5 to 10% of HCA, that did not display any specific morphological features previously described.

Based on this new classification, radiologists have rapidly shown that imaging, especially MRI, was accurate for the differentiation between HCA subtypes. Laumonier et al and Ronot et al described specific combinations of imaging features associated with each new subtype^26,27:

• H-HCAs are characterized by a diffuse and homogeneous signal dropout on chemical shift T ₁ weighted sequences. Using only this feature, the sensitivity of MRI ranges from 87 to 91% and the specificity ranges from 89 to 100% for diagnosing H-HCA,^26,27

• I-HCAs are characterized on MRI by a strong hyperintense signal on T ₂ weighted images, marked APHE, and persistent enhancement on delayed phase using extracellular MR contrast agents. Using only two features (signal intensity on T ₂ and on delayed phase) the sensitivity of MRI ranges from 85 to 88% and the specificity ranges from 88 to 100% for diagnosing H-HCA^26,27 (Figure 3).

• The two other subtypes show far less specific features. Therefore, differentiation between b-HCA from unclassified HCA and HCC was not possible.^26,27

Figure 3.

A 36-year-old female with inflammatory adenoma of the right liver lobe. Contras-enhanced MR with extracellular contrast agents was performed for the characterization of a fortuitously discovered focal liver lesion. The patient had no medical history. T ₂ weighted images showed a heterogeneous lobulated lesion with high signal intensity (arrow in A). The lesion showed arterial phase hyperenhancement (arrow in B) with contrast retention on delayed phase images (arrow in C). Pathological analysis with hematoxylin eosin staining showed sheets and cords 1–3 cells thick of normal appearing hepatocytes, sinusoidal dilatation (D–E) explaining delayed contrast retention, and inflammatory infiltrates (E). The high number of impaired arterial vessels, explaining the arterial phase hyperenhancement of the tumor.

During the following years, several studies have been published refining the radiological appearance of the different HCA subtypes, but also further subdividing HCA categories by identifying new mutations or abnormal signalling pathways. For instance, exome sequencing analysis identified other b-catenin mutations in exons 7 and 8 in HCA that were previously recognized as unclassified or I-HCA subgroups. It also separated sonic-hegdehog HCA (sh-HCA) from the group of unclassified tumors. The result is a complex diagnostic classification linking each of the different HCA subtypes to specific epidemiology, pathophysiology, natural history, and management.²⁸

As stated above, imaging has also benefitted from the introduction of liver-specific MR contrast agents. Indeed, on HBP, FNH have been shown to be associated with iso- or hyperintensity whereas HCA typically appear hypointense.²⁹ As a consequence, hepatobiliary MR contrast agents were shown to be very useful for the differentiation between FNH and HCA. Yet, several studies have reported HCAs, especially I-HCA, showing paradoxical iso- or hyperintensity on HBP in 26–67% of cases.³⁰ The study of Ba-Ssalamah et al, therefore reported a sensitivity and specificity of 80.9 and 77.3%, respectively, using gadoxetic acid for the diagnosis of the I-HCA subtype, lower than that reported with extracellular MR contrast agents.³¹ This is in contradiction to the molecular background of I-HCA, as their OATP expression has been shown to be lower than that of the adjacent liver.³² The explanation probably lies in the surrounding liver parenchyma, which frequently displays marked steatosis that modifies the relative signal intensity of tumors. On the contrary, molecular studies have shown that OATP expression is persistent in b-HCA.³² In the study of Ba-Ssalamah et al, five out of the six b-HCAs displayed retention of gadoxetic acid on the hepatobiliary phase.³¹

This second example based on the history of HCA illustrates a second type of rad-path relationship: pathological references that are constantly changing and improving lead to continuous radiological progress.

When imaging surpasses pathology

In the first two examples, pathology was and has remained the undisputed reference technique, either fixed (HCC), or changing (HCA). Another possible, but rarer situation is encountered when imaging surpasses its pathological reference, becoming the new reference. Imaging of hepatic steatosis will help us understand this.

The example of hepatic steatosis

Hepatic steatosis corresponds to triglyceride storage in the liver, as a response to various causes. In the absence of competing liver disease etiology, particularly excessive alcohol consumption, hepatic steatosis is considered to be the first step towards the development of non-alcoholic fatty liver disease (NAFLD), i.e. the liver component of the metabolic syndrome. This syndrome is a cluster of metabolic disorders (visceral obesity, insulin resistance, hypertension, hypertriglyceridemia and low HDL-cholesterol), and is associated with an increased risk of developing cardiovascular diseases and Type 2 diabetes. A recent meta-analysis including 8,515,431 patients from 22 countries reported a global prevalence of NAFLD of 25%.^33–35 Hepatic steatosis may evolve into non-alcoholic steatohepatitis (NASH), and cirrhosis, with an increased risk of HCC. NAFLD is expected to become the leading cause of chronic liver disease. The diagnosis of steatosis is therefore very important.

The definition is pathological, hepatic steatosis being present when > 5% of hepatocytes contain lipid droplets.³⁶ Pathologists frequently use semi-quantitative scores, classifying steatosis as mild (5–33%), moderate (33–66%), and severe forms (>66%).³⁷ Different imaging techniques have been evaluated for the non-invasive diagnosis of steatosis. On ultrasound, liver hyperechogenicity and organs and vessels shading have been described. Webb et al also proposed a quantitative measurement based on the echogenicity of the right liver and the renal cortex.³⁸ On CT, liver attenuation has been shown to be lower in steatotic livers.

³⁹ Yet, the diagnostic performance of these imaging techniques has been shown to be promising but insufficient, especially for the identification of mild to moderate steatosis.⁴⁰

MRI, on the other hand, is based on nuclear magnetic resonance (NMR). This principle is used in NMR spectroscopy to obtain physical, chemical, electronic and structural information about molecules. Following the same principle, proton MR spectroscopy has been shown to accurately detect and quantify the absolute amount of fat in the liver.^41–43 Yet, it is time-consuming, not widely available, and requires spectral analysis skills. This is why it is not in routine use.

Chemical shift-based MR sequences rely on the difference between the resonant frequency of protons linked to water and fat. Acquisition of in and out-phase gradient echo T ₁ weighted images provides a simple quantification of the proton density fat fraction (PDFF, in %) (Figure 4). It is a great deal easier and faster than ¹H-MR spectroscopy. However, it is subject to multiple biases (including, but not limited to, the T ₁ effect, the influence of the flip angle, the T ₂* decay effect or the spectral complexity of hepatic lipids) that may significantly alter fat quantification. To compensate for all these biases, vendors have developed multiple gradient-echo sequences which involve the acquisition of multiple in- and out-phase images.⁴⁴ These breath-hold sequences are acquired in a matter of seconds. The MR signal is therefore fitted to a mathematical model taking into consideration all biases (Figure 4). These sequences have been shown to be strongly correlated to ¹H-MR spectroscopy or chemical lipid extraction^44–48 , and of course liver biopsy.^46,48,49 MR-PDFF has also been shown to be more reproducible than histological steatosis.⁴⁶ Yet, and importantly, since MR and biopsy don’t quantify the same thing, differences between histological steatosis and MR-PDFF have been shown, PDFF being roughly half that reported by pathologists. The MRI-PDFF threshold for the diagnosis of steatosis is around 6.5%, close to the first pathological definition of liver steatosis (triglycerides accounting for >5% of liver volume) published by Hoyumpa in 1975.⁵⁰ As a consequence, gastroenterologists and hepatologists have suggested that it is time to replace assessment of liver histology with MR-based imaging tests to assess the efficacy of interventions for NAFLD,⁵¹ and to use MR-PDFF as a new gold-standard for the detection and grading of steatosis (Figure 5).⁵² Of course, quantification of hepatic fat using MR doesn’t solve all problems, because of availability and cost of MRI scanners, and the need for both technical and reader expertise. This is why, radiology stimulating radiology, numerous published and ongoing studies have tried to develop non-invasive ultrasound-based techniques (attenuation imaging, quantification of speed of sound, etc.) that are performed much more easily in daily practice.⁵³

Figure 4.

A 47-year-old female with marked steatosis assessed by MRI. On gradient echo T ₁ weighted images, the liver showed marked and homogeneous signal dropout on out-of-phase (A) when compared with in-phase (B) sequence, corresponding to marked steatosis. Multiple gradient echo T ₁ weighted sequences allowed for the non-invasive quantification of the fat content. On the PDFF map, a region of interest was drawn, and showed 53.4% PDFF. PDFF, proton density fat fraction.

Figure 5.

Illustration of the differences between pathological and MR imaging-based quantification of hepatic steatosis. MRI is an accurate, precise and reader-independent non-invasive imaging quantitative biomarker of liver fat content, capable of steatosis quantification of the entire liver, while pathology provides a more reader-dependent semi-quantitative assessment performed on a tissue sample.

This third and final example illustrates another kind of rad-path relationship, when radiology surpasses pathology and creates a paradigm shift.

Authors have captured the multidisciplinary convergence between radiology, pathology, and genomics, as exemplified by the present paper, by the term “integrated diagnostics”.⁵⁴ Such collaboration is anticipated to deeply integrate the workflow between these different disciplines. The obstacles are not technological, but rather the traditional approaches and silo organization.

Conclusion

This article humbly aimed at showing different ways in which radiology has evolved when facing different pathological references. When the reference is solid and stable over time (e.g. HCC), radiology is capable of correctly reaching a non-invasive diagnosis, using various combinations of imaging features and modalities, and creating complex diagnostic algorithms. When facing a changing reference (e.g. HCAs), radiology is constantly challenged, and rapidly improves. Finally, in rare situations, radiology is capable of surpassing pathological definitions (e.g. hepatic steatosis), using advanced imaging modalities. We showed that not only did pathology and radiology grow for decades, but that their level of interminglement grows ever more, to the point that their definitive fusion is to be anticipated sooner than later, moving imaging into the larger field of integrated diagnostics.