Project

Development and Validation of a Machine Learning Algorithm to Predict Negative-Margin (R0) Resectability of Pancreatic Head Adenocarcinoma based on Preoperative CT Scan Images

Code
1260123N
Duration
01 November 2022 → 31 October 2025
Funding
Research Foundation - Flanders (FWO)
Research disciplines
  • Medical and health sciences
    • Computational biomodelling and machine learning
    • Data visualisation and high-throughput image analysis
    • Development of bioinformatics software, tools and databases
    • Abdominal surgery
    • Oncological surgery
Keywords
Machine learning-based predictive model Implementing and prospective validation Predicting R0 resectability of pancreas tumor
 
Project description

Pancreatic ductal adenocarcinoma has the highest mortality rate of all major cancers with a five-year relative survival rate of 9%. Despite progress in preoperative imaging and neoadjuvant treatment, the incidence of microscopically positive margins (R1) after the Whipple procedure is as high as 30-40%, and incomplete resection is associated with poor survival. The current standard of care is to use preoperative CT scan to assess resectability, but the sensitivity and specificity in detecting vascular involvement are low at 41-77% and 81%, respectively. Thus, there is a critical need for better preoperative assessment of anticipated margin status. Artificial intelligence, specifically machine learning (ML) and computer vision, is a potential solution to overcome this issue. These technologies grant us the ability to classify imaging based on patterns not seen by the human eye and have been shown to outperform experts in diagnosing multiple cancers, including breast, brain, and skin. The objective of this project is to develop and validate an algorithm that improves the preoperative prediction of margin status with an accuracy of >80% among patients with resectable or borderline resectable pancreas head cancer, regardless of neoadjuvant chemotherapy status. Using this model, clinicians can more effectively and selectively identify patients with a high likelihood of R1, and subsequently offer therapeutic adjuncts to improve their chances of achieving a negative margin.