Abstract
AbstractIntroductionPancreatic cancer is thought to have an extremely dismal prognosis. Most cancer-related deaths occur from metastasis rather than the primary tumor, although individuals with tumors smaller than 1 cm in diameter have more than 80% 5-year survival. Thus, the current protocol introduces PanCanAID project which intends to develop several computer-aided-diagnosis (CAD) systems to enhance pancreatic cancer diagnosis and management using CT scan imaging.Methods and analysisPatients with pathologically confirmed pancreatic ductal adenocarcinoma (PDAC) or pancreatic neuroendocrine tumor (PNET) will be included as pancreatic cancer cases. The controls will be patients without CT evidence of abdominal malignancy. A data bank of contrast-enhanced abdominopelvic CT scans, survival data, and demographics will be collected from ten medical centers in four provinces. Endosonography images and clinical data, if available, will be added to the data bank. Annotation and manual segmentation will be handled by radiologists and confirmed by a second expert radiologist in abdominal imaging. PanCanAID intelligent system is designed to (1) detect abdominopelvic CT scan phase, (2) segment pancreas organ, (3) diagnose pancreatic cancer and its subtype in arterial phase CT scan, (4) diagnose pancreatic cancer and its subtype in non-contrast CT scan, (5) carry out prognosis (TNM stage and survival) based on arterial phase CT scan, (6) and estimate tumor resectability. A domain adaptation step will be handled to use online data and provide pancreas organ segmentation to reduce the segmentation time. After data collection, a state-of-the-art deep learning algorithm will be developed for each task and benchmarked against rival models.ConclusionPanCanAID is a large-scale, multidisciplinary AI project to assist clinicians in diagnosing and managing pancreas cancer. Here, we present the PanCanAID protocol to assure the quality and replicability of our models. In our experience, the effort to prepare a detailed protocol facilitates a positive interdisciplinary culture and the preemptive identification of errors before they occur.
Publisher
Cold Spring Harbor Laboratory