Automatic Segmentation of Pancreas and Pancreatic Tumor: A Review of a Decade of Research

Abstract

In the current era of machine learning and radiomics, one of the challenges is the automatic segmentation of organs and tumors. Tumor detection is mostly based on a radiologist's manual reading, which necessitates a high level of professional abilities and clinical experience. Moreover, increasing the high volume of images makes radiologists' assessments more challenging. Artificial intelligence (AI) can assist clinicians in diagnosing cancer at an early stage by providing a solution for assisted medical image analysis. The automated segmentation of tumor is better realized through conventional segmentation methods and, nowadays, through machine learning and deep learning techniques. The segmentation of abdominal organs and tumors from various imaging modalities has gained much attention in recent years. Among these, pancreas and pancreatic tumor are the most challenging to segment and have recently drawn a lot of attraction. The main objective of this paper is to give a summary of different automated approaches for the segmentation of pancreas and pancreatic tumors and to perform a comparative analysis using various indices such as dice similarity coefficient (DSC), sensitivity (SI), specificity (SP), precision (Pr), recall and Jaccard index (JI), etc. Finally, the limitations and future research perspectives of pancreas and tumor segmentation are summarized.