IPMN-LEARN: A linear support vector machine learning model for predicting low-grade intraductal papillary mucinous neoplasms

Backgrounds/Aims

We aimed to build a machine learning tool to help predict low-grade intraductal papillary mucinous neoplasms (IPMNs) in order to avoid unnecessary surgical resection. IPMNs are precursors to pancreatic cancer. Surgical resection remains the only recognized treatment for IPMNs yet carries some risks of morbidity and potential mortality. Existing clinical guidelines are imperfect in distinguishing low-risk cysts from high-risk cysts that warrant resection.

Methods

We built a linear support vector machine (SVM) learning model using a prospectively maintained surgical database of patients with resected IPMNs. Input variables included 18 demographic, clinical, and imaging characteristics. The outcome variable was the presence of low-grade or high-grade IPMN based on post-operative pathology results. Data were divided into a training/validation set and a testing set at a ratio of 4:1. Receiver operating characteristics analysis was used to assess classification performance.

Results

A total of 575 patients with resected IPMNs were identified. Of them, 53.4% had low-grade disease on final pathology. After classifier training and testing, a linear SVM-based model (IPMN-LEARN) was applied on the validation set. It achieved an accuracy of 77.4%, with a positive predictive value of 83%, a specificity of 72%, and a sensitivity of 83% in predicting low-grade disease in patients with IPMN. The model predicted low-grade lesions with an area under the curve of 0.82.

Conclusions

A linear SVM learning model can identify low-grade IPMNs with good sensitivity and specificity. It may be used as a complement to existing guidelines to identify patients who could avoid unnecessary surgical resection.

Evaluating triage protocols for endoscopic ultrasound-guided fine needle biopsies of the pancreas

INTRODUCTION

Pancreatic endoscopic ultrasound-guided fine needle biopsy (EUS-FNB) acquires both direct smear and small core biopsy specimens. The triage protocols for pancreatic FNBs to cytopathology (CP) or gastrointestinal surgical pathology (GIP) are controversial and vary by institution.

MATERIAL AND METHODS

Pancreatic EUS-FNBs obtained with the SharkCore FNB were reviewed from January 2014 to June 2019. The specimen characteristics and pathology data, including tissue triage, were obtained from the electronic medical records. We assessed the diagnostic yield, defined as malignant, specific neoplastic, or benign, and the operating characteristics at the time of rapid on-site evaluation (ROSE) and final diagnosis.

RESULTS

We reviewed 324 pancreatic FNBs from 313 patients. Of the 324 FNBs, 260 (80%) obtained concurrent direct smear and core biopsy specimens, 30 (12%) of which were divided between CP and GIP. Of the 51 core-only specimens, 47 (92%) were reviewed by CP and 4 (8%) by GIP. ROSE improved the overall diagnostic yield by 10% and accuracy by 9%. When core specimens were reviewed independently, the diagnostic accuracy was 93% for CP (n = 248) and 100% for GIP (n = 33). All false-negative results of the CP-reviewed cores were due to sampling error. Concurrent smear review improved EUS-FNB performance, increasing the negative predictive value by 10% and accuracy by 3% compared with core review alone.

CONCLUSIONS

CP and GIP can accurately interpret pancreatic EUS-FNB specimens. However, triage of concurrent EUS-FNB-acquired smear and core specimens to CP may be most efficient as CPs are trained to assess adequacy at the time of ROSE, as well as interpret all parts of the biopsy, minimizing the risk of discordant pathology reports.

Diagnostic yield of the SharkCore EUS-guided fine-needle biopsy

INTRODUCTION

Endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA) is the standard diagnostic procedure for many intrathoracic and intra-abdominal lesions. Next-generation fine-needle biopsies (FNBs) can increase diagnostic yield by procuring tissue suitable for histological processing. We evaluate the diagnostic yield and operating characteristics of the SharkCore (SC; Medtronic Corp., Minneapolis, MN) FNB in a tertiary referral facility.

MATERIALS AND METHODS

We performed a single-center retrospective review of SC-FNB-acquired tissue between January 2014 and March 2018. Patient demographic data, endoscopic features, and pathology data were obtained from the electronic medical record. Diagnostic yield was assessed by the ability to obtain a definitive diagnosis, defined as malignant or benign interpretations. Operating characteristics were also calculated.

RESULTS

A total of 179 lesions were sampled with the SC-FNB in 157 patients (mean age: 63 years, 57% male). Of these, 31 lesions were concomitantly sampled with a conventional FNA needle. Most lesions were pancreatic (49%). Diagnostic yield was 86%, which was independent of lesion location, lesion size and needle gauge. Diagnostic accuracy was highest when both histology and cytology specimens were analyzed concurrently (96.5%). In patients with a history of chronic pancreatitis, accuracy, sensitivity, and negative predictive value were reduced (71.4%, 20.0%, and 69.2%, respectively). Rapid onsite evaluation (ROSE) occurred in 64.8% of cases and was more likely to be diagnostic at the time of rapid evaluation if SC-acquired tissue was utilized versus FNA-acquired tissue (P = 0.03); however, final diagnostic yield did not differ between needles (P = 0.13).

CONCLUSIONS

SC-FNB shows high diagnostic yield and accuracy and provides diagnostic tissue for ROSE. SC-FNB is an effective alternative to conventional FNA.