Management of Patients With Pancreatic Cysts: Analysis of Possible False-Negative Cases of Malignancy

Visual Analytics of a Computer-Aided Diagnosis System for Pancreatic Lesions

Machine learning is a powerful and effective tool for medical image analysis to perform computer-aided diagnosis (CAD). Having great potential in improving the accuracy of a diagnosis, CAD systems are often analyzed in terms of the final accuracy, leading to a limited understanding of the internal decision process, impossibility to gain insights, and ultimately to skepticism from clinicians. We present a visual analytics approach to uncover the decision-making process of a CAD system for classifying pancreatic cystic lesions. This CAD algorithm consists of two distinct components: random forest (RF), which classifies a set of predefined features, including demographic features, and a convolutional neural network (CNN), which analyzes radiological (imaging) features of the lesions. We study the class probabilities generated by the RF and the semantical meaning of the features learned by the CNN. We also use an eye tracker to better understand which radiological features are particularly useful for a radiologist to make a diagnosis and to quantitatively compare with the features that lead the CNN to its final classification decision. Additionally, we evaluate the effects and benefits of supplying the CAD system with a case-based visual aid in a second-reader setting.

Abdominal Imaging of Pancreatic Cysts and Cyst-Associated Pancreatic Cancer in BRCA1/2 Mutation Carriers: A Retrospective Cross-Sectional Study

BACKGROUND

Direct-to-consumer BRCA testing will increase BRCA diagnoses and subsequent abdominal imaging. It is unclear whether BRCA carriers are at higher risk of developing pancreatic cysts (PCs) or cyst-associated pancreatic ductal adenocarcinoma (PDAC). We investigated the prevalence of PCs in BRCA-tested patients, and whether BRCA-carriers have higher rates of PDAC when PCs are found.

STUDY DESIGN

This is a retrospective cross-sectional study of patients with BRCA testing and abdominal imaging between 1996 and 2018. Pancreatic cysts were identified on original imaging reports. Prevalence and risk characteristics of PCs, as well as incidence of PDAC, were compared between BRCA+, BRCA-, and BRCA-untested patients.

RESULTS

Pancreatic cysts were identified in 4,045 patients among 128,164 unique patients with abdominal imaging, including 33 patients with PCs in 1,113 BRCA-tested patients. There was no difference in PC prevalence between BRCA+, BRCA-, and untested patients (3.6%, 2.6%, 3.2%, respectively; p = 0.64). Pancreatic cysts were diagnosed in BRCA+ patients at a younger age (57.1 vs 65.3 years, p < 0.001); however, there was no difference in risk stratification compared with BRCA- or untested patients by consensus criteria. Across the population of imaged patients, patients with PCs had significantly higher rates of PDAC compared with those without PCs (18.2% vs 2.4%, p < 0.001). Incidence of cyst-associated PDAC was similar in BRCA+ and BRCA- patients (13.3% vs 22.2%, p = 0.84).

CONCLUSIONS

BRCA+ patients have similar rates of PCs, high-risk features in their cysts, and PDAC as BRCA- and untested patients. BRCA+ patients likely do not require dedicated abdominal imaging to evaluate for PCs and should follow management guidelines similar to those as the untested general population if an incidental PC is identified.

Genetic testing vs microforceps biopsy in pancreatic cysts: Systematic review and meta-analysis

BACKGROUND

Carcinoembryonic antigen (CEA) and cytology in pancreatic cystic fluid are suboptimal for evaluation of pancreatic cystic neoplasms. Genetic testing and microforceps biopsy are promising tools for pre-operative diagnostic improvement but comparative performance of both methods is unknown.

AIM

To compare the accuracy of genetic testing and microforceps biopsy in pancreatic cysts referred for surgery.

METHODS

We performed a literature search in Medline, Scopus, and Web of Science for studies evaluating genetic testing of cystic fluid and microforceps biopsy of pancreatic cysts, with endoscopic ultrasound with fine-needle aspiration (EUS-FNA) prior to surgery and surgical pathology as reference standard for diagnosis. We evaluated the diagnostic accuracy for: 1- benign cysts; 2- mucinous low-risk cysts; 3- high-risk cysts, and the diagnostic yield and rate of correctly identified cysts with microforceps biopsy and molecular analysis. We also assessed publication bias, heterogeneity, and study quality.

RESULTS

Eight studies, including 1206 patients, of which 203 (17%) referred for surgery who met the inclusion criteria were analyzed in the systematic review, and seven studies were included in the meta-analysis. Genetic testing and microforceps biopsies were identical for diagnosis of benign cysts. Molecular analysis was superior for diagnosis of both low and high-risk mucinous cysts, with sensitivities of 0.89 (95%CI: 0.79-0.95) and 0.57 (95%CI: 0.42-0.71), specificities of 0.88 (95%CI: 0.75-0.95) and 0.88 (95%CI: 0.80-0.93) and AUC of 0.9555 and 0.92, respectively. The diagnostic yield was higher in microforceps biopsies than in genetic analysis (0.73 vs 0.54, respectively) but the rates of correctly identified cysts were identical (0.73 with 95%CI: 0.62-0.82 vs 0.71 with 95%CI: 0.49-0.86, respectively).

CONCLUSION

Genetic testing and microforceps biopsies are useful second tests, with identical results in benign pancreatic cysts. Genetic analysis performs better for low- and high-risk cysts but has lower diagnostic yield.

Should pancreas cyst fluids be divided into two for cytological diagnosis and biochemical tests?

BACKGROUND/AIMS

The aim of the present study was to investigate whether pancreas cyst fluids should be divided into two for cytological diagnosis and biochemical tests.

MATERIALS AND METHODS

The present study was conducted with fluids aspirated from 12 pancreas cysts. The fluids were divided into two and sent to the cytopathology (fluid 1) and biochemistry (fluid 2) laboratories. Fluid 1 was centrifuged at the cytopathology laboratory. Cytology slides were prepared from the deposit, and the supernatant was sent to the biochemistry laboratory. Fluid 2 was centrifuged at the biochemistry laboratory, and amylase, carcinoembryonic antigen, and cancer antigen 19.9 levels were determined in the supernatant. These procedures were repeated for fluid 1 from the cytopathology laboratory. The remaining fluid 2 was sent to the cytopathology laboratory. Fluid 1-like slides were prepared from fluid 2 in the cytopathology laboratory. Cytological diagnoses of fluid 1 and fluid 2 were compared, and the Pearson correlation coefficient for biochemical test results was identified.

RESULTS

92% of fluid 1 and 50% of fluid 2 were diagnostic. Biochemical test results of fluid 1 and fluid 2 were similar, and the Pearson correlation coefficient was high.

CONCLUSION

Our results showed that pancreatic cyst fluids did not need to be divided into two for cytological diagnosis and biochemical tests. Following centrifugation of the whole fluid at the cytopathology laboratory, the deposit and the supernatant can be used for cytological diagnosis and for biochemical tests, respectively. With this protocol, the sensitivity of cytological diagnoses and biochemical tests of pancreatic cyst fluids may increase.

Incremental value of DNA analysis in pancreatic cysts stratified by clinical risk factors

BACKGROUND AND AIMS

We determined the incremental predictive value of pancreatic cyst fluid molecular analysis to assessing malignancy risk over long-term follow-up of a well-characterized cohort, given the underlying predictive value of imaging parameters routinely used to triage such patients.

METHODS

Patients who lacked initial cytologic malignancy in cyst fluid and had final pathology or a follow-up period of more than 2 years were included. Patient outcomes determined the malignancy-free survival of patients with high-risk stigmata (HRS), worrisome features (WFs), and DNA abnormalities. DNA analysis included 3 abnormalities: loss of heterozygosity mutations among a panel of tumor suppressor genes, Kras mutation, and elevated DNA quantity.

RESULTS

Included were 478 patients; 209 had surgical pathology-derived outcomes and 269 had clinical follow-up of >2 years. Eleven percent had malignant outcome. Forty-two patients had HRS, 272 lacked both HRS and WFs, and 164 lacked HRS but had WFs. DNA abnormalities did not statistically change long-term malignancy risk in patients with HRS or in patients lacking both HRS and WFs. Among patients with WFs, the presence of ≥2 DNA abnormalities significantly increased malignancy risk (relative risk, 5.2; P = .002) and the absence of all DNA abnormalities significantly decreased risk (relative risk, .4; P = .040). Sensitivity analysis confirmed results of survival analysis over differing baseline malignancy probabilities.

CONCLUSIONS

Our study defines the clinical characteristic of patients in which DNA abnormality testing has the greatest impact on patient outcomes. Use of DNA abnormality testing is supported in a carefully selected patient population limited to cysts with WFs.