Predicting Pancreatic Ductal Adenocarcinoma (PDAC) Through Artificial Intelligence Analysis of Pre-Diagnostic CT Images

点击翻译按钮获取中文摘要

The objective of the proposed project is to develop a Pancreatic Ductal Adenocarcinoma (PDAC) prediction model to identify individuals who have high risk for PDAC in the next 3 years through Artificial Intelligence (AI) analysis of pre-diagnostic CT images and non-imaging factors. PDAC is the fourth leading cause of cancer- related deaths in both men and women in the United States despite its low incidence rate. The 5-year survival rate for all stages of PDAC is 10% but can be as high as 50% with early-stage diagnosis. Therefore, identification of individuals at high risk for PDAC has high clinical significance as follow-up imaging examinations or biopsy may assist in early detection and allow surgical intervention while the tumors are still resectable. However, PDAC prediction is difficult due to the lack of reliable screening tools, the absence of sensitive and specific symptoms and biomarkers, and low prevalence. Abdominal pain is the single most common reason that Americans visit the emergency room (ER), where an abdominal Computed Tomography (CT) scan is usually performed. Even though most scans don’t show any signs of cancer visible to the naked eyes of radiologists, some subjects eventually develop PDAC in the next few years. These pre-diagnostic CT images provide critical morphological information associated with biological changes at the pre-cancer or early cancer stage, which can be extracted using AI to predict PDAC risk. Therefore, the objective of the proposed project is to uncover unique features in pre-diagnostic images using AI and develop PDAC prediction model based on these features. Non-imaging factors such as demographic, epidemiologic, and anthropometric factors, clinical comorbidities, and laboratory tests will be included in the model to improve the prediction accuracy. The primary hypotheses are a) AI allows extraction of unique image features in pre-diagnostic CT images associated with pre-cancer or early cancer biological changes that are invisible to naked eyes and b) the combination of pre-diagnostic image features and non- imaging factors improves the accuracy of PDAC risk stratification and prediction over that using conventional non-imaging factors alone. To verify these hypotheses, we will retrospectively evaluate CT pancreatic images obtained up to 3 years prior to PDAC diagnosis that were deemed non-cancerous by radiologists. A group of subjects who underwent similar imaging studies for non-gastrointestinal disorders and were age/gender matched with pre-diagnostic imaging will serve as healthy controls. Accurately stratifying high risk individuals may allow for early detection of PDAC in the future. A major challenge of the project is the scarcity of the appropriate imaging data because of the low prevalence of PDAC and stringent enrollment criteria. Eight major medical centers will participate in collection of 1,064 cases. The end point of this project is the development, training, and validation of an AI-based PDAC prediction model, which will identify individuals who are at high risk for developing PDAC within the next 3 years.