AI for Good

Chapter 28Using Artificial Intelligence to Inform Pancreatic Cyst Management

—Juan M. Lavista Ferres, Felipe Oviedo, William B. Weeks, Elliot Fishman, and Anne Marie Lennon

Executive Summary

Pancreatic cancer is the most lethal cancer, marked by a distressingly low five-year survival rate of just over 12 percent. Over the past decades, while many cancer types have seen advancements in treatment and prognosis, pancreatic cancer continues to be tantamount to a death sentence for many patients.

A proportion of pancreatic cancer cases originate from pancreatic cysts—abnormal masses in the pancreas. While these cysts are detectable and potentially removable, their surgical removal is challenging: while the surgery's complexity and associated risks vary case by case, there is a high mortality rate in certain scenarios. Further, not all pancreatic cysts progress to cancerous states; cysts that are not pre-cancerous require much less aggressive treatment. Given the high prevalence of pancreatic cystic lesions—recent studies show that 17 percent of individuals in their 30s and over 75 percent of those older than 80 have pancreatic cysts—accurately identifying the nature of pancreatic cysts could improve care outcomes, reduce treatment risks, and reduce healthcare costs.

Our study explored the potential of an explainable boosting machine (EBM) learning model. This EBM model, trained on past surgical cases, significantly surpassed clinical guidance in discerning the cyst type. It provides ...