Anupama Ramachandran, Kumble Seetharama Madhusudhan

Abstract Gastroenteropancreatic neuroendocrine neoplasms comprise a heterogeneous group of tumors that differ in their pathogenesis, hormonal syndromes produced,biological behavior and consequently, in their requirement for and/or response to specific chemotherapeutic agents and molecular targeted therapies. Various imaging techniques are available for functional and morphological evaluation of these neoplasms and the selection of investigations performed in each patient should be customized to the clinical question. Also, with the increased availability of cross sectional imaging, these neoplasms are increasingly being detected incidentally in routine radiology practice. This article is a review of the various imaging modalities currently used in the evaluation of neuroendocrine neoplasms, along with a discussion of the role of advanced imaging techniques and a glimpse into the newer imaging horizons, mostly in the research stage.

Key Words: Neuroendocrine tumor; Gastroenteropancreatic; Intravoxel incoherent motion;Diffusion weighted imaging; Perfusion imaging; Dual energy computed tomography

lNTRODUCTlON

Gastroenteropancreatic neuroendocrine neoplasms (GEPNENs) are a rare and heterogeneous group of tumors that originate from the gastrointestinal and pancreatic neuroendocrine cells[1,2]. They may be benign or malignant and may or may not secrete hormones. Use of the previous terminology - carcinoid tumor - is no longer encouraged. The gastrointestinal tract (GIT), having the highest density of neuroendocrine cells in the body, is the most common site of involvement of neuroendocrine neoplasms (NENs),comprising nearly 60% of all NENs[3]. Pancreatic NENs (PNENs) account for about 7% of all GEPNENs[4]. Most PNENs are sporadic. However, association with four familial syndromes (in up to 25%) is well described: multiple endocrine neoplasia type I, von Hippel Lindau syndrome, neurofibromatosis type I and tuberous sclerosis[5,6].

Classification based on histological differentiation and grade is desirable as it provides insight into tumor biology, clinical course and helps in planning management. In 2019, the 5thedition of the World Health Organization (WHO) classification of tumors series published the latest NEN classification(Table 1)[7]. The latest WHO classification recognizes that well-differentiated NENs may be high grade,but they are distinct from the poorly differentiated neuroendocrine carcinomas (NECs). Most GEPNENs are well-differentiated and slowly growing. Grade 3 NENs are most common in the pancreas, but can occur throughout the GIT[7]. Given the differences in prognosis, tumor grade is the most important factor determining the treatment of GEPNENs. The treatment of GEPNENs depends on grade, differentiation, site of origin, and stage of tumor, and the opinion about the best treatment strategy is evolving.Surgical resection remains the cornerstone and is the only curative treatment. For patients with small (＜2 cm), low-grade NENs, decisions on surgeryvsactive surveillance need to be individualized based on tumor size, morphology (homogeneous, well circumscribed tumor ＜ 1 cm correlate with low malignant potential) and patient characteristics like age and presence of comorbidities[8].

The commonly used imaging modalities include ultrasonography (US), computed tomography (CT),magnetic resonance imaging (MRI) and positron emission tomography (PET)-CT. Imaging is primarily aimed at accurate detection, characterization and staging of these neoplasms and also at assessment of response to treatment. Sensitivity of common imaging modalities used in the evaluation of GEPNENs are summarized in Table 2.

Improvements and advances in the imaging techniques have mainly focused on the noninvasive prediction of the grade of the NENs. The European Neuroendocrine Tumor Society (ENETS) has also recommended that preoperative assessment of the grade of the NENs is essential for prognosis prediction and management planning[9]. In addition, with the improvements in the imaging technologies, the detection rates of small NENs have significantly improved, with many often being detected incidentally.

US

Transabdominal ultrasonography (USG) often is the commonest initial modality used for patients with gastrointestinal symptoms. It has value in the detection of liver metastases (sensitivity reaching 85%-90%). PNENs in general appear as hypoechoic masses with a hyperechoic halo on USG[10].

However, transabdominal USG has limitations. It has a poor sensitivity (13%-27%) for the detection of GEPNENs[11]. The technique is dependent on the experience of the operator. Presence of bowel gas and increased subcutaneous fat can obscure adequate visualization.

The advent of harmonic imaging, pulse inversion sequence, low mechanical index techniques and ultrasound contrast agents (UCAs) has enabled routine application of contrast-enhanced ultrasound(CEUS) to overcome the limitations of conventional B mode USG. The inherent advantage of CEUS is its ability to assess tumor-enhancement patterns in real time during transabdominal USG[12]. The enhancement patterns are described during arterial, portal venous and late phases. UCAs are gas microbubbles stabilized by a shell, the composition of which varies depending on the type of contrast agent. UCAs are blood pool agents and increase the back scatter of US, enhancing the echogenicity of flowing blood[12]. Harmonic imaging detects harmonic signals from the microbubbles and CEUS specific US modes filter signals from the background tissue, thereby showing even very slow blood flow without Doppler related artifacts. UCAs, unlike CT and MRI contrast agents, are excreted by the lungs and hence can be used safely in patients with deranged renal function.

Table 1 World Health Organization Classification and grading criteria for neuroendocrine neoplasms of the gastrointestinal tract and hepatopancreatobiliary organs (7)

Table 2 Sensitivity of common imaging modalities used in the evaluation of gastroenteropancreatic neuroendocrine neoplasms

The differential perfusion on CEUS has been shown to identify and diagnose pancreatic tumors.Pancreatic adenocarcinomas are in general hypovascular, while NENs are hypervascular. Takedaet al[13] found three patterns of hyperenhancement of PNENs and found that CEUS was useful in the differentiation of PNENs from pancreatic adenocarcinomas. Malagòet al[14] also showed that the enhancement patterns of nonfunctioning PNENs on CEUS (hyper-, iso- or hypovascular) correlated well with Ki-67 index and CEUS improved the detection of hepatic metastases. Hypervascular lesions had lower Ki-67 index. Another study showed that the enhancement patterns of NENs on CEUS correlated significantly with CT enhancement pattern and histological Ki-67 index and CEUS was a good predictor of response of tumors to somatostatin analogues[15].

Elastography

Elastography is an advancement in USG that enables real-time measurement of tissue stiffness along with display in colors superimposed on the grey scale images[16]. In general, elastography helps with the differentiation of benign and malignant lesions based on stiffness, as malignant lesions are usually hard. Only a few studies have reported the usefulness of transabdominal shear wave elastography(SWE) in the evaluation of pancreatic tumors. Parket al[17] showed that elastography can differentiate benign and malignant solid pancreatic lesions based on the difference in the shear wave velocity values(relative stiffness) between the tumor and the normal parenchyma. An early study by Uchidaet al[18]found that NENs were homogeneous and soft on elastography, comparable to the normal pancreas.They also reported that a combination of elastography and B mode USG, improved the diagnostic accuracy to 90%, from 70%-80 % when B mode USG was used alone. However, if the visualization on the baseline B mode USG is suboptimal, the results of CEUS and elastography are also often unsatisfactory. These limitations have been overcome by the use of endoscopic ultrasound (EUS).

EUS

EUS is considered the most accurate test for the diagnosis of pancreatic masses[19]. EUS uses higher frequency (7.5-12 MHz) probes, placed in proximity to the area of interest and hence performs better at detection of tumors ＜ 2 cm for which CT and MRI have poor sensitivity[20]. Overall, EUS has a sensitivity of 82%-93% and a specificity of 92%-95% for localizing PNENs. EUS is particularly useful in the detection of benign insulinomas that lack somatostatin receptors and consequently are not detected on somatostatin receptor scintigraphy/single photon emission computed tomography (SPECT)/PET[21]. EUS also plays an important role in the detection of functional pancreatic and extrapancreatic(duodenal) gastrinomas[22]; both of which generally have a small size (average 1 cm) at diagnosis. The additional benefit of EUS is its ability to guide accurate tissue samplingviafine needle aspiration and core biopsy[23]. CEUS can also be performed through EUS with the use of second generation UCAs (e.g,Sonovue), which produce harmonic signals at low acoustic powers. A recent study showed that the time intensity curve analysis during CEUS showed high diagnostic accuracy in grading PNENs, and could differentiate grade 1/2 tumors from grade 3 tumors/carcinomas[24].

EUS elastography

EUS elastography of the pancreas has been shown to be a promising imaging technique in several studies[25-27]. However, a prospective study by Hircheet al[28] including 70 patients with undifferentiated pancreatic masses showed an overall lower sensitivity (41%), specificity (53%) and accuracy (45%)for the detection of malignancy. One of the early studies evaluating conventional strain elastography for pancreatic lesions showed significant strain difference between benign and malignant masses[29]. The major utility of EUS elastography is in increasing the yield of sampling by aiding better tumor targeting,especially in the background of pancreatic parenchymal fibrosis[30]. SWE EUS is a recent development and studies evaluating its utility in the pancreas are just emerging[31,32]. A recent comparative study suggested that conventional strain elastography was superior to SWE in the characterization of pancreatic lesions[33].

CT

CT is the cornerstone and the most commonly performed imaging modality for the diagnosis and preoperative staging of GEPNEN. Standard CT scan has low sensitivity (～60%) for the detection of GIT NENs[34]. Dynamic dual-phase protocol, which includes the arterial and portal venous phases, is recommended in patients with suspected NEN[35,36] (Figure 1). For PNEN, the late arterial phase acquired at 40-45 s (pancreatic phase) suits best. For detection of small bowel NENs, CT enterography(oral administration of neutral contrast agent like mannitol for bowel distension) or CT enteroclysis(administration of neutral contrastvianasojejunal tube) is required. CT enteroclysis combines the advantages of enteroclysis with imaging capabilities of multidetector computed tomography (MDCT).Study by Kamaouiet al[37] showed that CT enteroclysis had 100% sensitivity and 96.2% specificity for the detection of small bowel NEN. A comparative study from Mayo Clinic found that CT enterography was better than capsule endoscopy in detecting small bowel tumors, with a sensitivity of up to 93%[38].This ability of CT enterography to detect small bowel tumors remains high (sensitivity of 88%) even in the presence of gastrointestinal bleeding[39].

The characteristic imaging feature on CT scan suggesting the diagnosis of NENs is their intense enhancement in the late arterial phase, owing to the hypervascular nature of the tumor. The arterial phase also helps in outlining the relationship of the tumor with the adjacent arteries. Using the maximum intensity projection technique, virtual angiographic images can be obtained. Volume rendering techniques applied to the arterial phase provide easily explainable images to the surgeon.Portal venous phase helps to draw the relationship of the tumor with major veins, especially the splenic vein and the superior mesenteric vein for PNENs. Dual-phase imaging is also crucial for the evaluation of hepatic metastases. The classic liver metastases from NEN, being hypervascular, are most evident on the arterial phase images. About 6%-15% of NEN liver metastases are appreciated only in the arterial phase[40,41]. However up to 16% are hypovascular and show delayed enhancement[40].

One of the early studies suggested that the size of the tumor is an important prognostic factor, with tumors ＜ 1 cm showing lesser incidence of liver metastasis (20%-30%) compared to those with ＞ 1 cm (＞40% risk)[42]. Studies have shown that up to 42% of PNENs may not show arterial phase hyperenhancement[43,44]. Such arterial phase hypoenhancing tumors were associated with a significantly lower 5-year survival (54%) compared to the lesions which were isoenhancing (89%) or hyperenhancing (93%)[45]. Rodallecet al[46] found that tumor enhancement on CT scan correlated with microvascular density(MVD) evaluated on histology and that hypoenhancing PNENs correlated with poorly differentiated tumors and a decrease in overall survival rate. These studies showed that the CT enhancement characteristics of NENs have a prognostic value. Gallottiet al[6] found that incidentally detected PNENs, size＞ 3 cm, complex enhancement pattern and, presence of calcification, vascular invasion, main pancreatic duct dilatation, and peripancreatic lymph nodes were associated with nonbenign tumors and required more aggressive course of management.

Figure 1 49-year-old man with epigastric pain and raised serum gastrin levels. A and B: Axial contrast enhanced pancreatic phase computed tomography (CT) images show a well-defined hyperenhancing mass (arrow in A) in the head and neck of pancreas, abutting the proper hepatic artery along with two hyperenhancing focal lesions in the liver (arrowheads in B), indicating hepatic metastases; C and D: Axial portal venous phase CT images show retention of contrast in the lesions in both locations. Thickened gastric mucosal folds is also noted (arrowheads in C).

The role of various CT quantitative parameters based on enhancement of the NEN in the arterial or pancreatic and the venous phases has been evaluated in the prediction of tumor grade. Kimet al[47]found that portal enhancement ratio (HU value of the tumor divided by the HU value of pancreatic parenchyma on portal phase images) had the best odds ratio (49.6) and a cutoff value of ＜ 1.1 had a sensitivity of 92% and specificity of 81% in differentiating grade 3 PNENs from grade 1 and 2. This high sensitivity and specificity of portal enhancement ratio in differentiating neuroendocrine carcinomas from well-differentiated NENs was also confirmed by another study[48]. Yamadaet al[49] showed that corrected true enhancement values in the pancreatic phase had a sensitivity of 92%, specificity of 84%and area under the curve of 0.897 in the differentiation of grade 1 from grade 2 PNENs. D’Onofrioet al[50] showed that various tumor enhancement parameters (tumor permeability ratios, tumor parenchyma ratios, tumor arterial ratio and tumor venous ratio) were significantly different between grade 1 and grade 3 and between grade 2 and grade 3 PNENs. However, these values could not differentiate grade 1 from grade 2 tumors.

Dual-energy CT

Dual-energy CT (DECT) is an advancement in CT, which allows acquisition of images at two energy levels, with lower energy being 80-100 kVp and higher being 140 kVp. Using DECT, material decomposition of images and generation of iodine maps, virtual noncontrast images, and monochromatic images at different energy levels is possible (Figures 2 and 3). Monoenergetic images at low keV (55 keV) in the pancreatic phase of DECT show improved image contrast for evaluation of pancreatic masses[51]. Monochromatic spectral images improve the sensitivity of detection of NENs like insulinomas, particularly the hypovascular and isoattenuating tumors and the sensitivity is comparable to MRI[52]. One study showed that iodine uptake obtained from DECT is useful in the differentiation of hepatocellular carcinoma from liver metastases arising from NENs, with the former showing significantly higher iodine uptake (3.8 ± 1.2vs2.3 ± 0.6)[53]. This iodine uptake parameter on DECT may also be used in assessing the response to treatment of NENs.

Perfusion CT

Figure 2 Dual-energy computed tomography images of a 46-year-old man presenting with melena. A: Axial monochromatic computed tomography(CT) image at 55 keV in pancreatic phase shows a hyperenhancing well-defined mass (arrow) arising from the duodenal wall; B: Enhancement of the same lesion(arrow) appears subtle on the axial 100 keV monochromatic CT image; C and D: Iodine overlay maps show bright areas (arrow) suggesting contrast uptake (C), with iodine concentration of 3.2 mg/mL in areas of uptake (D). Iodine concentration of normal pancreas was 0.5 mg/mL.

Figure 3 Dual-energy computed tomography in grading the pancreatic neuroendocrine neoplasms. A: Iodine overlay dual-energy computed tomography (CT) map of a 40-year-old woman with low-grade (grade 2) pancreatic neuroendocrine neoplasms (PNEN) in head of pancreas shows hyperenhancement of the tumor with an iodine concentration of 5.1 mg/mL; B and C: Iodine overlay dual energy CT maps of a 29-year-old man with grade 3 PNEN(outlined in B) shows large hypoenhancing areas with low iodine concentration (0.9 mg/mL) and peripheral bright areas with iodine concentration of 4.3 mg/mL (C).Measuring iodine concentration helps in objectively assessing the grade of the tumor.

Perfusion CT is a technique that measures the dynamic changes in the attenuation of the tissues after contrast administration. It allows quantitative measurement of tissue perfusion, thereby assisting in the assessment of tumor viability and biological behavior[54]. The commonly used quantitative parameters of perfusion CT in oncoimaging are blood flow, blood volume, vascular permeability-surface area product and mean transit time (Figures 4 and 5). These parameters serve as imaging biomarkers of tumor angiogenesis, which is ideally assessed histologically by calculating the MVD[55]. NENs are among the tumors with significant angiogenesis. Unlike majority of the cancers, where increased tumor vascularity is associated with aggressive behavior, higher microvascular density in NENs is associated with a low tumor grade[54]. Low MVD was found to be an unfavorable prognostic factor for PNENs in several studies despite the presence of other favorable conventional histoprognostic factors, and call for a more aggressive treatment approach[56-58]. A study by d’Assignieset al[59] on 36 patients with PNENs found a significant correlation between MVD and blood flow assessed by using perfusion CT. In their study, the authors found that tumors that are small (＜ 2 cm), benign (grade 1), with a proliferation index of ≤ 2%, and without histological signs of microvascular involvement had a significantly higher blood flow. Volume perfusion CT has been shown to improve the detection of pancreatic insulinomas,particularly the ones which have transient hyperenhancement (comprising 30% cases)[60,61]. A recent study demonstrated that addition of low dose perfusion CT to contrast enhanced CT improved the detection rate of PNENs from 83.6% to 89.1% and found that blood flow parameters were significantly different between grade 1 and grade 2 tumors[62].

Perfusion CT has also been shown to have a role in monitoring response to treatment with antiangiogenic drugs. Few studies have shown that the perfusion parameters of PNENs and liver metastases decrease as early as 48 h after treatment with anti-angiogenic drugs and perfusion CT offers a significant role in an early noninvasive assessment[63,64]. A major limitation of perfusion CT is the higher radiation dose, resulting in an additional dose of approximately 7 mSv[65].

Figure 4 Volume perfusion computed tomography images of a 56-year-old man with recurrent hyperinsulinemic hypoglycemia. A: Axial pancreatic phase computed tomography image shows a hyperenhancing lesion in the pancreatic tail (arrow); B-D: Color-coded parametric maps for blood volume (B),blood flow (C) and mean transit time (D) of the tumor (arrow) and normal pancreatic tissue; E: Chart shows mean value of each perfusion parameter of the tumor.Blood flow in the tumor was higher (247 mL/100 mL/min) compared to normal pancreatic parenchyma (72 mL/100 mL/min); F: Time attenuation curve shows dynamic enhancement pattern of the tumor corresponding to transient hyperenhancement. Histopathology after enucleation proved the tumor to be grade 1 insulinoma.

Figure 5 Volume perfusion computed tomography images of a 67-year-old man with a large grade 3 neuroendocrine neoplasm involving body and tail of pancreas. A and B: Axial arterial phase computed tomography images with circular regions of interest placed at two different locations in the lesion (*); C-H: Parametric maps for blood flow (C and D) and blood volume (E and F) with mean value of each perfusion parameter (G and H) are shown. Lower values of mean blood flow, mean blood volume and mean transit time are features of high grade neuroendocrine neoplasm.

MRl

MRI is best performed as a problem-solving tool when CT scan findings are equivocal or negative, and is aimed at acquiring images of the lesion and organ with better soft tissue contrast. For instance, MRI has shown better sensitivity for the detection of liver metastases compared to CT and somatostatin receptor scintigraphy[66]. The absence of exposure to ionizing radiation makes MRI the apt modality for screening young individuals suspected of having NEN and those with syndromic association who require multiple follow up imaging[67]. Most NENs are hypointense on T1-weighted and hyperintense on T2-weighted images[10]. Contrast enhancement pattern and morphologic appearances are similar to that seen in CT scan (Figure 6). For the detection of PNENs, the sensitivity of MRI ranges from 85% to 100% and specificity from 75% to 100%[68].

Diffusion-weighted imaging

Diffusion-weighted imaging (DWI) is a widely used technique in clinical imaging as it reflects the microscopic environment of the neoplasm including tumor cellularity and extracellular matrix. The application of DWI in oncology is mainly in tumor detection and assessing response to chemotherapy and radiotherapy. Wanget al[69] demonstrated that the apparent diffusion coefficient (ADC) values of PNEN correlated well with Ki-67 labelling index, thus indicating that DWI has a prognostic value(Figure 6). Another study showed that the ADC values were significantly different between benign and non-benign PNENs (1.48 × 10-3mm2/svs1.04 × 10-3mm2/s, respectively)[70]. Lotfalizadehet al[71]showed that DWI has the additional value in identification of high grade tumors (grade 3) and can accurately differentiate grade 3 from grade 1/2 tumors (AUROC-0.96). The ADC values showed an inverse relation with the grade of the tumor.

Another major utility of DWI is in the detection and characterization of liver metastases. Several studies have shown that DWI is more sensitive for the detection of liver metastases than T2-weighted and multiphase gadolinium-enhanced MRI, especially for smaller lesions[72-74]. Besaet al[75] showed that the ADC of liver metastases from NEN weakly and significantly correlated negatively with tumor grade and Ki-67 and that the mean ADC and the minimum ADC values were significantly different between the three grades (1.6, 1.35 and 0.9 × 103mm2/s and 0.84, 0.5 and 0.27 × 103mm2/s for grades 1, 2 and 3, respectively). DWI is hence recommended in routine MRI abdomen protocol for the detection of liver metastases from NEN.

Histogram analysis of the ADC of the whole tumor has also been shown to predict tumor grade and aggressiveness. Pereiraet al[76] found that whole tumor histogram analysis of the ADC, including the skewness and kurtosis can reliably differentiate grade 1 from grade 2/3 tumors. Another study also showed that this histogram analysis of ADC was useful in predicting tumor grade, vascular invasion and metastasis (node, liver) in PNENs and that ADCentropyand ADCkurtosiswere the best markers in identifying tumor aggressiveness[77].

DWI is also useful in predicting and assessing response to various medical treatments for NENs. A recent study by Le Bihanet al[78] showed that the change in the ADC values of liver metastases from NENs after transarterial radioembolization was significantly different between the partial response and progressive disease groups, thus concluding that ADC can be used as an additional marker for treatment response evaluation.

While DWI investigates diffusion of water molecules in tissues, it does not detect perfusion of blood.Intravoxel incoherent motion (IVIM) DWI detects translational motion of water molecules in a voxel and can simultaneously quantify their diffusion and microcirculation in tissue capillary network[79]. IVIM images are quantified by ADC, which integrates the effects of both diffusion and perfusion. IVIM therefore enables evaluation of tissue perfusion without the requirement of a contrast agent. The quantitative parameters in IVIM include the pure diffusion coefficient (Dslow), which reflects the diffusion of water molecules, the pseudodiffusion coefficient (Dfast), which reflects the diffusion movement of capillary microcirculation perfusion, and the perfusion fraction (f), which represents the volume ratio between the perfusion effect of local microcirculation and the overall molecular diffusion(Figures 7 and 8). IVIM-DWI is a useful method to assess true tumor cellularity of PNEN, represented by tissue diffusion, as increased microcirculation of hypervascular PNENs may cause the pseudodiffusion effect and thus leads to the overestimation of ADC values[79]. Hwanget al[80] observed that IVIM DWI can differentiate grade 1 from grade 2 or 3 PNENs. They found that pure diffusion coefficient is a better marker of tumor cellularity than ADC, and was significantly higher in grade 1 PNENs,thereby enabling prediction of tumor grade on imaging. A recent study showed that Dslowand Dfastparameters help in the differentiation of high grade PNENs from pancreatic adenocarcinoma with high diagnostic accuracy (0.460vs0.579 × 103mm2/s and 13.361vs4.985 × 103mm2/s, respectively)[81].

Diffusion kurtosis imaging

Diffusion kurtosis imaging (DKI) is a new rapidly advancing MRI technique based on the concept that water molecules in biological environment have non Gaussian properties. This is in contrast to standard DWI that calculates ADC using monoexponential analysis, assuming that diffusion of water in tissues follows Gaussian behavior[82]. At higher b values (＞ 1000 s/mm2), due to the barriers encountered by water molecules in tissues, there is deviation from Gaussian distribution. The deviation when quantified, in fact represents the tissue microenvironment. Two quantitative parameters, diffusion coefficient (D) and kurtosis (K), representing deviation from Gaussian distribution, can be extracted from DKI. DKI thus provides a more accurate model of diffusion and quantifies tissue heterogeneity,and irregularity of cellular microstructure by capturing non Gaussian diffusion parameters (Figure 9)[82,83]. A drawback which hampers its use in routine practice is the long acquisition time due to scan acquisition at multiple b values. There are studies showing application of DKI for the assessment of the pancreas[84,85]. Shiet al[83] found that the radiomics model of DKI and T2 weighted imaging could improve the diagnostic accuracy for PNENs.

Figure 6 Magnetic resonance images of a 24-year-old woman with multiple endocrine neoplasia-type 1 syndrome and pancreatic neuroendocrine neoplasm. A-C: Axial magnetic resonance images through the head of pancreas show a round heterogeneous mass (arrow) which appears hypointense on T1-weighted image (A), hyperintense on T2-weighted image with central cystic / necrotic change (B) and shows peripheral hypointensity on apparent diffusion coefficient (ADC) image (C) suggesting diffusion restriction along the periphery (ADC = 0.93 × 103 mm2/s); D-F: Axial dynamic contrast enhanced T1-weighted images show hyperenhancement of the tumor along the periphery in pancreatic phase (D), with contrast retention in venous (E) and delayed (F) phase images. The patient also had bilateral inferior parathyroid and left superior parathyroid adenomas.

Figure 7 lntravoxel incoherent motion diffusion-weighted imaging in a 46-year-old woman with proven grade 1 pancreatic neuroendocrine neoplasm. A: Axial diffusion weighted image (b = 800 s/mm2) shows a small pancreatic lesion with diffusion restriction (arrow); B: Color-coded diffusion map shows true diffusion coefficient, D = 2.33 × 103 mm2/s; C: Color-coded perfusion map shows pseudodiffusion coefficient, D* = 5.37 × 103 mm2/s); D: Color-coded perfusion fraction map shows a value, f = 3.2%; E: Signal decay curve of the tumor (purple) shows fall in signal at lower b values with plateau at higher b values. In comparison, normal pancreas (orange) shows lesser diffusion restriction than the tumor.

MR elastography

MR elastography (MRE) is a phase-contrast-based MRI technique for the evaluation of mechanical tissue properties noninvasively,e.g.,tissue stiffness. MRE of pancreas is at an early stage. Recent studies have used MRE to differentiate healthy from pathological pancreatic tissue[86]. The normal pancreatic stiffness in adults measured by MRE is 1.1-1.21 kPa[86]. Shiet al[87] used MRE for the characterization of solid pancreatic masses. They found that malignant masses had significantly higher stiffness (3.27 kPa) than benign masses (1.96 kPa). PNENs had a median stiffness of 2.32 kPa. They also suggested that stiffness ratio (ratio of stiffness of mass to normal parenchyma) may perform better in the differentiation of benign from malignant pancreatic masses.

Figure 8 lntravoxel incoherent motion diffusion-weighted imaging in a 67-year-old man with grade 3 pancreatic neuroendocrine neoplasm. A: Axial diffusion weighted image (b = 200 s/mm2) with circular region of interest in the tumor; B: Color-coded diffusion map shows true diffusion coefficient, D = 0.84 × 103 mm2/s; C: Color-coded perfusion map shows pseudodiffusion coefficient, D* = 5.01 × 103 mm2/s); D: Color-coded perfusion fraction map shows a value, f = 4.5%; E: Signal decay curve shows steeper decay at low b values and continued fall at higher b values [in comparison to grade 1 pancreatic neuroendocrine neoplasm (PNEN) (Figure 7), true diffusion coefficient is lower in grade 3 PNENs].

Figure 9 Diffusion kurtosis of the lesion same as in Figure 8. A: Axial diffusion-weighted image (b = 200 s/mm2) with region of interest marked; B:Diffusion map (D = 3 × 103 mm2/s); C: Kurtosis map (k = 0.40); D: Signal decay curve shows non-Gaussian diffusion.

MRI perfusion

MRI perfusion techniques for the assessment of tumor perfusion have the major advantage that they lack adverse effects of radiation compared to the radiation-intensive CT perfusion. T1-weighted dynamic contrast enhanced (DCE) MRI is the technique applied in the evaluation of tumors in the abdomen[88]. This provides both semiquantitative and quantitative information on the microvascular perfusion of the tissue. The semiquantitative analysis is based on the time-signal intensity curve and the quantitative analysis is based on the Tofts two-compartment pharmacokinetic model (intravascular and extravascular-extracellular compartments) with the parameters evaluated being Ktrans (volume transfer constant, wash in), Kep (reverse efflux rate constant, wash out) and Ve (extravascular extracellular space volume fraction)[89]. This technique has shown promising results in the evaluation of PNENs. A study by Donatiet al[90] showed that Ktrans and Kep values were higher in NENs (2.709 ± 0.110/min; 5.957 ±0.371/min) compared to other focal lesions and healthy pancreatic parenchyma. This in fact reflects the wash in (Ktrans) and wash out (Kep) of the contrast agent from the hypervascular NENs. Also, well differentiated and poorly differentiated NENs showed different perfusion characteristics. In the study by Kimet al[91], Ktrans value of NENs were significantly higher than that of neuroendocrine carcinomas (0.339 ± 0.187/minvs0.077 ± 0.036/min). Ductal adenocarcinomas being hypovascular,show significantly lower average values of Ktrans and Kep[90-92]. The role of DCE MRI in the evaluation of response to systemic chemotherapy and targeted molecular therapy by assessing the changes in the values of MRI perfusion parameters reflecting good or poor response to treatment is a direction for future studies.

PET CT AND PET MRl

Hybrid anatomical and functional imaging using PET/CT is a valuable tool in the current practice of grading and management of NENs. In general, in dual tracer PET/CT (somatostatin receptor imaging with Ga68DOTATATE/TOC/NOC PET/CT and glucose metabolism with FDG PET/CT), low-grade tumors, which express somatostatin receptors, bind to somatostatin analog, but not to FDG[93]. In contrast, poorly differentiated GEPNENs with high Ki-67 index would be negative on somatostatin PET/CT but FDG avid[94]. Zhanget al[95] suggested that dual tracer PET /CT may be used as an alternative to tissue sampling, as it reflects both cellular somatostatin receptor expression and glucose metabolism. The authors also found a positive correlation between SUVmax (standardized uptake value) and Ki-67 index with respect to FDG PET/CT and negative correlation with respect to Ga68DOTATATE PET/CT. However, low-grade insulinomas show low expression of somatostatin receptors in contrast to other secreting and nonsecreting NENs and are frequently not detected on Ga68DOTATATE PET/CT[96]. Since virtually all benign insulinomas express glucagon-like peptide 1 (GLP-1) receptors (incretin receptors), these receptors can be targeted by PET/CT for preoperative localization of occult benign insulinomas. GLP-1R PET/CT had higher sensitivity than MRI and SPECT/CT for localization of benign insulinomas in a study by Antwiet al[97]. Glucose dependent insulinotropic polypeptide receptor (GIPR) is another incretin receptor overexpressed in GEPNENs. It is a potential target for imaging the small percentage (～10%) of GEPNENs which do not express SSTR and GLP1R, as confirmed by studies in animal models[98].

With technical advances, simultaneous PET and MRI acquisition in an integrated scanner is now possible. The first study in 2013 by Beiderwellenet al[99] showed that every lesion detected on PET/CT was identified on PET/MRI. Hopeet al[100] evaluated hepatic lesions in patients with NENs using Ga68DOTATOC PET/CT and PET/MRI and found that there was a strong correlation between SUVmax obtained in PET/CT and PET/MRI. However, due to the high cost, the routine use of PET/MRI is limited.

RADlOMlCS, TEXTURE ANALYSlS AND MACHlNE LEARNlNG

Radiomics is the process of conversion of digital biomedical images to mineable data and the subsequent analysis of this data[101]. Texture analysis is an imaging technique under the wider arena of radiomics, that extracts, analyzes and interprets quantitative imaging features, and enables objective assessment of tumor heterogeneity beyond what is possible to human eyes[102]. In statistical-based model of texture analysis, from each voxel in a region of interest, various first order (e.g.,first-order entropy, kurtosis, skewness, standard deviation, mean intensity) and second order (e.g.,contrast,uniformity, second order entropy,etc.), or higher order features are extracted and analyzed using post processing software. As mentioned previously, tumor grade is an important prognostic factor of NENs and their prediction noninvasively is valuable.

CT radiomics is increasingly finding its place in the grading of NENs. Canellaset al[103], evaluating PNENs on CT scan, found that tumors with high entropy (a texture parameter reflecting tissue heterogeneity) values had 3.7 times higher odds of being aggressive (grades 2 and 3). In this study, entropy was a better predictor of tumor grade than the size of the lesion. Choiet al[104] found that lower kurtosis, lower sphericity and higher skewness correlated to grade 2 or 3 PNENs. A study on 3D texture analysis of PNENs in 100 patients showed that kurtosis was significantly different between all the three grades and entropy could differentiate grade 1 from grade 3 and grade 2 from grade 3, but not grade 1 from grade 2 tumors[50]. These results of CT texture analysis were confirmed in other recent studies[105,106] thus emphasizing its role in the prediction of tumor grade.

MRI radiomics also help in characterizing PNENs. MRI texture analysis was found to be useful in differentiating nonfunctioning PNEN from solid pseudopapillary neoplasm in a study by Liet al[107].Nonlinear discriminant analysis was found to have the lowest misclassification rate of all the types of analyses performed in their study. Shindoet al[108] studied ADC histogram for differentiation of pancreatic adenocarcinoma from PNENs. In their study, ADC entropy had the highest area under the curve (AUC) for differentiating adenocarcinoma from NEN. De Robertiset al[77] found that ADChistogram analysis of DWI, using radiomics, could predict aggressiveness of PNENs. They found high ADC kurtosis values in tumors with vascular invasion (AUC of 0.763 for a cut off value of 4.13) and distant metastases (AUC of 0.820 for a cut off of 3.642)[77]. The future prospects of radiomics are in the direction of development of a robust predictive model combining qualitative and quantitative imaging parameters.

Table 3 Summary of important research studies on imaging of gastroenteropancreatic neuroendocrine neoplasms

EUS: Enhanced ultrasonography; TIC: Time-signal intensity curve; APCT: Arterial phase computed tomography; ADC: Apparent diffusion coefficient;CECT: Contrast enhanced computed tomography; CEMRI: Contrast enhanced magnetic resonance imaging; CNN: Convolutional neural network; DWI:Diffusion weighted imaging; NEC: Neuroendocrine carcinoma; NEN: Neuroendocrine neoplasm; AER: Annual equivalent ratio; PER: Portal enhancement ratio; MVD: Microvascular density; WHO: World Health Organization; MDCT: Multidetector row computed tomography; OS: Overall survival; PFS:Progression free survival; AUC: Area under the curve; PNEN: Pancreatic neuroendocrine neoplasm; GAN: Generative Adversarial Network.

Machine learning is increasingly being used in medicine and has various applications including detection of disease, classification of images, identifying treatment and monitoring adherence to therapy[109,110]. The standard radiomics analysis on CT or MRI requires marking of the tumor margins for analysis. However, deep learning using convolutional neural network (CNN) performs analysis automatically and provides better results[111]. A few recent preliminary studies have shown the promising role of deep learning using CNN in the prediction of grade of PNEN and survival using contrast enhanced CT[112-114]. Clinical trials for translation of these imaging techniques into clinical practice and validation for routine use are ongoing.

In short, the quantitative parameters derived from imaging, relevant for prognostication of GEPNENs include tumor size, enhancement ratios derived from HU values, iodine uptake on DECT, entropy on CT texture analysis, tumor blood flow, tumor blood volume, and mean transit time on perfusion CT,ADC and ADC histogram analysis of DWI, true and pseudodiffusion coefficients and perfusion fraction on IVIM DWI, Ktrans and Kep on perfusion MRI and SUVmax on dual tracer PET/CT. A combination of qualitative features and quantitative factors with the newer functional imaging techniques enables better tumor classification based on their prognosis. A summary of important studies on advanced imaging in GEPNENs is shown in Table 3.

CONCLUSlON

With the recent advances in CT, MRI, USG and hybrid imaging techniques like PET/CT and PET/MRI using dual tracers, smaller pancreatic and bowel NENs are now being increasingly detected and staged.In addition to tumor detection and staging, their non-invasive grading, prognostication and monitoring response to treatment are shown to be feasible and reliable with the emerging studies using quantitative imaging techniques like CT and MR perfusion studies, DWI, IVIM and texture analysis with radiomics.Standardization of these techniques with more large scale studies would be an important future prospect. These advances in imaging will help in making the right treatment choice, contributing to an overall improvement in patient outcome.

FOOTNOTES

Author contributions:Ramachandran A contributed to initial draft and revision; Madhusudhan KS contributed to review of draft and revision.

Conflict-of-interest statement:No conflict of interests.

Open-Access:This article is an open-access article that was selected by an in-house editor and fully peer-reviewed by external reviewers. It is distributed in accordance with the Creative Commons Attribution NonCommercial (CC BYNC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is noncommercial. See: https://creativecommons.org/Licenses/by-nc/4.0/

Country/Territory of origin:India

ORClD number:Anupama Ramachandran 0000-0002-5808-6076; Kumble Seetharama Madhusudhan 0000-0001-8806-2625.

Corresponding Author's Membership in Professional Societies:Indian Radiological and Imaging Association, No.501LM/D-285A.

S-Editor:Zhang H

L-Editor:Kerr C

P-Editor:Zhang H