PANCREAS AND PANCREATIC TUMOR SEGMENTATION USING AI: A COMPREHENSIVE REVIEW OF DEEP LEARNING INNOVATIONS

Abstract
Pancreatic cancer is known to be one of the most fatal cancers because of
the late detection of the disease, which is caused by anatomical
complexity and inconspicuous appearance on imaging. Manual
interpretation of radiological data can be very lengthy and it can be extremely
variable. The paper gives an in-depth description of the advances that occurred in
the previous decade of the research topic of pancreas and pancreatic tumor
segmentation and the corresponding use of knowledge through enhancement of
image processing algorithms to deep learning architectures such as U-Net, Attention
U-Net, and transformer-based architectures. We classify segmentation techniques as
conventional, and supervised, and unsupervised techniques and compare their
results based on a coefficient of similarity such as Dice Similarity Coefficient (DSC),
Jaccard Index (JI), sensitivity, specificity and precision. Moreover, we discuss recent
problems, publicly available datasets, and future research opportunities with respect
to multimodal learning, explainability, and standard evaluations on the clinical
deployment.