Potential role of imaging for assessing acute pancreatitis-induced acute kidney injury

Abstract

Acute kidney injury (AKI) is a common complication of acute pancreatitis (AP) that is associated with increased mortality. Conventional assessment of AKI is based on changes in serum creatinine concentration and urinary output. However, these examinations have limited accuracy and sensitivity for the diagnosis of early-stage AKI. This review summarizes current evidence on the use of advanced imaging approaches and artificial intelligence (AI) for the early prediction and diagnosis of AKI in patients with AP. CT scores, CT post-processing technology, Doppler ultrasound, and AI technology provide increasingly valuable information for the diagnosis of AP-induced AKI. Magnetic resonance imaging (MRI) also has potential for the evaluation of AP-induced AKI. For the accurate diagnosis of early-stage AP-induced AKI, more studies are needed that use these new techniques and that use AI in combination with advanced imaging technologies.

Introduction

Acute pancreatitis (AP) is a common inflammatory condition that is associated with an increased risk of remote organ complications and mortality. Severe AP (SAP) is associated with multi organ failure and poor prognosis. The kidneys are among the most commonly affected organs in patients with AP, and acute kidney injury (AKI) has long been recognized as a common and severe complication of AP. ¹ AKI in AP patients reduces the survival rate and increases the healthcare burden. Therefore, early prediction and diagnosis of AKI in AP patients is essential for proper management and improving long-term survival.

Currently, the diagnosis and staging of AKI are based on the 2012 KDIGO guidelines and the occurrence of one of the following criteria ² :

• An increase of serum creatinine (SCr) by at least 0.3 mg dl⁻¹ (26.4 μmol L⁻¹) within 48 h.

• An increase of SCr to at least 1.5-times above the baseline within 7 days.

• Urine volume less than 0.5 mL kg⁻¹ h⁻¹ for 6 h.

However, 6 h urine output is often difficult to accurately measure in clinical practice because of errors in measurement of time and because of urine retention in the drainage tube or bladder. ³ In addition, urine output monitoring often necessitates urinary bladder catheterization, and this can lead to infection. ⁴ Although the SCr level can be used to estimate glomerular filtration function, these levels are influenced by multiple factors, such as age, gender, muscle metabolism, protein intake, fluid balance, and drug effects. ⁵ Previous studies suggested that the SCr level may not change until the glomerular filtration rate declines by 50%. ⁶ Therefore, SCr is an insensitive biomarker for the early diagnosis of AKI. Thus, exclusive reliance on SCr and urine volume for the diagnosis AKI will lead to missed diagnoses and delayed treatment for patients with mild disease, thereby leading to an increased risk of AKI and mortality. It is urgent to find more sensitive and accurate indicators to predict early-stage AKI.

Imaging plays an important role in the diagnosis and staging of AP, because it can help to establish the cause and identify complications. CT is currently the most widely used imaging modality for evaluating patients with suspected AP, because it provides rapid results and reliably predicts disease severity and outcome. ⁷ Although MRI in an acute care setting is not necessary for diagnosis, it could help to assess the severity of AP and identify local complications and soft tissue abnormalities. ⁸ Trans-abdominal ultrasound (US) is not used for the diagnosis of AP, although it is a cost-effective, non-invasive, and simple imaging modality that allows monitoring of kidney status and early detection of AKI. ⁹

The present review summarizes recent advances in the use of different imaging modalities for assessment of AKI and failure of other organs induced by AP, in an effort to improve the diagnosis, monitoring, and treatment of these patients.

CT for evaluation of AP-related organ failure

CT is one of the most used cross-sectional imaging modalities for the diagnosis of AP because the results provide evaluation of the extent of inflammation, the presence of pancreatic necrosis and local complications, and disease severity. Various CT-based scoring systems are used for evaluation of current status and prediction of prognosis in patients with AP. For example, the CT severity index (CTSI), which was developed in the early 1990s by Balthazar, can estimate the severity of AP. ¹⁰ This scoring system considers pancreatic and peripancreatic inflammation, and pancreatic parenchymal necrosis. A modified CTSI (MCTSI), formulated by Mortele et al. in 2004, is a simplified evaluation system that considers peripancreatic inflammation, the extent of pancreatic parenchymal necrosis, and extrapancreatic complications. ¹¹ In 2007, De Waele et al established the unenhanced extra pancreatic inflammation on CT (EPIC) score to assess extra pancreatic manifestations. ¹² Clinicians commonly use one or more of these scoring systems for evaluation of AP.

Many studies reported that CT scores correlated with classification of moderate AP to SAP (from the 2012 revised Atlanta classification), thus indicating the presence of transient or persistent organ failure. ^12–16 Other studies described the direct correlation of CT scores with pancreatitis-related organ failure. ^10–12,14 Two recent prospective studies ¹⁴ reported that the MCTSI score correlated with the severity of failure of remote organs (lung, heart, kidney, etc.) in AP. ¹⁴ Chen et al ¹⁵ found that the EPIC score predicted organ failure within 24   h of AP onset, but could not be used to determine the severity of disease or the number of failed organs.

Several studies compared the different imaging scores and also compared imaging scores with clinical scores in the prediction of organ failure in patients with AP. The clinical scoring systems examined include the Acute Physiology and Chronic Health Examination II (APACHE II) score, the bedside index of severity in AP (BISAP) score, the Ranson score, and several others. One study of 397 consecutive cases reported that CTSI score and MCTSI score each had a significant correlation with organ failure (p < 0.0001), and that these two CT indexes had no significant differences in prognostic performance. In addition, compared with the APACHE II score, the CTSI score and MCTSI score were more accurate in the diagnosis of SAP. ¹⁶ A prospective cohort study compared four clinical scoring systems in the prediction of SAP, and reported the area under curve (AUC) was 0.81 for the BISAP score, 0.94 for the Ranson score, 0.78 for the APACHE II score, and 0.84 for the CTSI score. ¹⁷ Another study by Harshit et al ¹⁸ found the AUCs of MCTSI for predicting SAP (0.919), organ failure (0.893), and ICU admission (0.993) were higher than the AUCs for the APACHE II score, BISAP score, and Ranson score. A study of 346 patients compared the abilities of seven CT scoring systems (CTSI score, MCTSI score, pancreatic size index (PSI) score, extrapancreatic (EP) score, EPIC score, mesenteric edema and peritoneal fluid (MOP) score, and Balthazar CT score) and two clinical scoring systems (APACHE II score and BISAP score), all determined at admission, on prediction of SAP. ¹⁹ The results indicated that the Balthazar CT score and CTSI score had the highest accuracy for predicting of AP severity; however, there were no statistically significant differences among these CT and clinical scoring systems. ¹⁹ A recent systemic meta-analysis showed that the AUC for prediction of the severity of AP was 0.80 for the CTSI score, 0.79 for the BISAP score, 0.83 for the MCTSI score, 0.73 for the C reactive protein (CRP) level, 0.81 for the Ranson score, and 0.80 for the APACHE II score, also with no statistically significant differences in accuracy. ²⁰ However, the CTSI score is an easily calculated and informative tool that remains valuable in routine clinical practice.

The above studies indicate that imaging scores have good accuracy in predicting organ failure in AP, and that there is no evidence for the superiority of any single scoring system. However, clinicians must also consider the usability and simplicity of the different scoring systems. Clinical scores and imaging scores each have advantages and disadvantages. For example, imaging scores can more directly indicate involvement of the pancreas and surrounding organs and tissues, and are simple to calculate; however, clinical scores can be repeatedly reevaluated in a short period of time. Thus, the combined use of imaging and clinical scores may be ideal. Future studies are needed to develop an innovative and comprehensive evaluation model or scale that has improved accuracy, practicability, and simplicity.

CT for predicting AP-induced AKI

Previous studies have reported correlations between CT scores and AP-related organ failure, and other imaging studies have examined AP-induced AKI. For example, the renal rim grade (RRG) score is a simple scoring system based on pararenal and perirenal inflammation whose predictive value is equivalent to other classical scoring systems. ²¹ The original RRG score did not distinguish bilateral perirenal and pararenal spaces in evaluation of inflammation. Wang, et al therefore developed a modified RRG (MRRG) score, which also considers these regions. The results indicated that scores from the MRRG, RRG, CTSI, MCTSI, EPIC, and the involvement of posterior pararenal space (PPS) within the first 48 h of onset were closely related to the occurrence of AKI and death in patients with SAP. For the prediction of AKI in SAP, the MRRG score had the highest AUC (0.9); PPS involvement was a potential prognostic factor for non-recovery of renal function in SAP patients. ²² Li et al ²³ found that the EPIC score had a good correlation with the APACHE II score and Ranson score, and provided a better prediction of AKI in SAP patients than the CTSI score and the Balthazar CT score. Their other important findings were that the renal cortex and medulla were not clearly demarcated in the arterial phase, which strongly suggested renal injury, and that capsulated inflammatory exudation predicted recovery from AKI. Notably, both MRRG and EPIC scores can be calculated using plain CT imaging, so administration of a potentially nephrotoxic contrast agent is not necessary. Therefore, the MRRG and EPIC scoring systems appear most suitable for evaluation of AP patients who have a risk for AKI and the results can be used as an early warning signal to clinicians about the need for special attention and appropriate intervention for such patients. However, these conclusions need further confirmation by large prospective studies.

CT post-processing technology for evaluation of AP-related organ failure

In 2015, Meyrignac et al ²⁴ determined the extent of intraperitoneal necrosis and exudate in 264 AP patients and then calculated the volume of extrapancreatic necrosis using imaging software. They found that the volume of extrapancreatic necrosis had a significant correlation with organ failure (p < 0.001) and that the optimal threshold volume for predicting SAP was 100 ml. For prediction of organ failure, their method had a sensitivity of 95% and a specificity of 83%; the corresponding values were 100 and 46% for the Balthazar score and 25 and 95% for the CTSI. Thus, the AUC of extrapancreatic necrosis was the highest among the scoring systems they compared. A subsequent study confirmed that extrapancreatic necrosis volume was better in predicting organ failure than other image-based scoring systems (Balthazar CT score, CTSI score, and MCTSI score) and was also better than a laboratory-based scoring system (CRP at 48–72 h after onset). ²⁵ A subsequent study found that the volume of acute fluid accumulation in the abdominal cavity of AP patients correlated with post-drainage organ failure. ²⁶ The acute fluid accumulation and necrosis are associated with an increased intra-abdominal pressure. An increased intra-abdominal pressure predicts a more severe course of AP and also contributes to the pathomechanism of AKI. ^27,28 Thus, these CT post-processing quantitative methods for analysis of abdominal acute fluid accumulation and necrosis in AP appear to have great value in prediction, diagnosis, and management AP-induced AKI, and are worthy of further study.

MRI for evaluation of AP-related organ failure

MRI is not routinely used for examination of patients with AP, but it is nonetheless a reliable diagnostic technique that provides multi parameter imaging, high-resolution images of soft tissues, and does not require ionizing radiation. MRI is valuable for AP patients who require subsequent examinations after the initial evaluation, such as assessment of organ failure. A study reported that the results from an MRI-based scoring system (Extrapancreatic Inflammation on Magnetic resonance imaging [EPIM]) had a strong correlation with EPIC score (r = 0.794, p < 0.001), and that the EPIM had an AUC of 0.844 for predicting SAP and an AUC of 0.817 for predicting organ failure, similar to the AUCs from the APACHE II and BISAP scores. The authors suggested that the EPIM score was better in assessing AP severity than the EPIC score, and was also better than the EPIC score in predicting organ failure in patients with early-stage AP. ²⁹

A 2017 study compared the accuracy of magnetic resonance elastography (MRE) and CECT for the prediction of AP severity. ³⁰ These researchers found that the accuracy of predicting AP based on pancreatic stiffness greater than 2.47 kPa was 85.5%, comparable to the accuracies of predicting AP from the CTSI score (75.0%), APACHE II score (88.2%), and BISAP score (78.9%). Pairwise comparisons indicated no significant differences after Bonferroni correction.

Very few studies have used MRI to assess AP, and MRI studies of AP-induced AKI are especially rare. Studies of other diseases reported that multi-parameter magnetic resonance imaging (MMRI) had diagnostic value for identification of AKI. However, clinical studies are still scanty, and most relevant studies examined animals. ^31–35

MMRI provides information on metabolism, blood supply, cell edema, and on other of tissues, in addition to morphological changes. Thus, MMRI may be able to detect early pathological changes in the kidneys before there is an elevation of SCr, and thus provide clinicians with early warning. Furthermore, as a radiation-free and contrast-free examination modality, MRI may have better value for AP patients with kidney damage or potential kidney damage. Further MRI studies of these patients are needed.

Doppler ultrasound for evaluation of AP-induced AKI

Resistance to blood flow occurs in patients with AKI and CKD, and clinicians, therefore, routinely use Doppler US during examinations to assess renal blood flow and diagnose vascular abnormalities. ^36,37 The renal resistance index (RRI), which is easily obtained by US, provides an accurate evaluation of renal blood flow status. Bellos et al ³⁸ analyzed the correlation between early changes in RRI and the extent of AKI in patients with AP. They showed that a high RRI correlated with an increased risk of severe AKI, and often indicated severe AKI progression. Ninet et al ³⁹ performed a meta-analysis of the predictive value of RRI for AKI and found that an elevated RRI was an independent predictor of progression to sustained AKI. A 2020 study compared the diagnostic value of RRI and a semiquantitative power Doppler US (PDU) score in the prediction of Stage 3 AKI in critically ill patients. ⁴⁰ The results indicated that PDU score accurately predicted Stage 3 AKI in patients with reduced cardiac index (CI) but not in patients with maintained CI. RRI was also a poor predictor of Stage 3 AKI in patients with reduced CI. Therefore, the value of RRI for the prediction of AP-induced AKI needs further evaluation. We conclude that although there is still some controversy, Doppler US has advantages because it is non-invasive, simple, and can be performed repeatedly to detect hemodynamic changes, predict renal damage, and monitor progress.

Prospect of artificial intelligence for evaluation

Artificial intelligence (AI) is playing an increasingly important role in medicine and healthcare, and some AI applications have focused on AP and AKI. For example, Tomašev et al ⁴¹ developed a deep recurrent neural network model to predict AKI. This model predicted 55.8% of all cases of AKI and 90.2% of cases who had subsequent need for dialysis. Thus, this model has the potential to become a valuable tool for clinical decision-making, and is worthy of further attention and verification.

Radiomics is an emerging technology that uses AI and other methods to extract quantitative information from clinical images. Radiomics provides analyses of image information that cannot be detected by radiologists alone, and complements the existing medical imaging knowledge system. The combined use of radiomics with modeling tools, statistical modeling, and machine learning, can assist clinical decision making. Although most radiomics studies have examined tumors, radiomics is also applicable to a variety of diseases that are characterized by increased inflammation. For example, Chen et al ⁴² found that a radiomics model based on CECT had promising power for predicting recurrent pancreatitis, and that this model provided important information to clinicians that allowed them to modify certain risk factors. Mashayekhi et al ⁴³ found the radiomic features on CECT imaging can differentiate patients with functional abdominal pain, recurrent AP, and chronic pancreatitis. Lin et al ⁴⁴ found that an MRI-based radiomics model performed well in the early prediction of AP severity and that the AUC of this radiomics model in the training group (0.917) and the verification group (0.848) were higher than the AUCs from the APACHE II, BISAP, and MRSI scores.

Although a small number of radiomics studies examining AP have appeared, there is a notable lack of studies using radiomics methods to evaluate AP-induced AKI. In principle, the use of radiomics methods with a larger number of patients could provide more reliable information about AP-induced AKI, and of its pathology and underlying genetic risk factors. More accurate and robust radiomics models could improve the accuracy of diagnosis and disease prediction, and reduce misdiagnoses that result from subjective evaluations. Radiomics has expanded the utility of medical imaging in clinical practice, and has great promise for personalized diagnosis and treatment. Therefore, this is a promising method for the evaluation of AP-related organ failure, including kidney damage, and for the early and precise prediction and diagnosis of this outcome.

Conclusions

AKI is a serious complication of AP and is associated with poor prognosis in these patients. Early identification and management of potential AKI patients will likely improve their outcomes. The two indicators currently used to diagnose AKI—urine output and serum creatinine—have limitations because of measurement inaccuracy and their hysteretic dynamics. Imaging techniques can be used for quantitative analysis of lesions, and functional parameters can reflect early changes in hemodynamics and tissue injury. CT scores, CT post-processing technology, Doppler US, and AI technology provide valuable information for the diagnosis of AP-induced AKI. MRI also has potential for the evaluation of AP-induced AKI. We believe that advanced imaging approaches, especially when combined with AI technology, can provide more accurate predictions of AP-induced AKI and also provide timely and accurate information to clinicians for scheduling different preventative strategies.