Deep Learning Techniques for Pancreatic Cancer Analysis: A Comprehensive Review of Imaging-based Approaches
Year: 2026
Type: conference-paper
Abstract: Pancreatic cancer accounts for 3% of all cancer cases but 7 % of all cancer related deaths with a five year survival rate of less than 10 % making it one of the most lethal cancers. This review summarizes the findings of 11 recent studies published in the last five years on the pancreatic cancer detection and segmentation techniques that utilized deep learning methods on CT and MRI datasets. Each study evaluated different types of models (including CNNs, Vits, Hybrid Architectures, etc.) and performance metrics. Detection performance of all studies was found between$85 \%-95 \%$accuracy, segmentation performance using Dice score was between$\mathbf{7 3 \% - 9 6 \%}$. The studies did, however, exhibit a number of limitations including that 9 out of 11 studies used single center dataset and only 2 performed external validation which raises the concerns about the generalizability of the models. Furthermore, the majority of transformer-type models required between$8-16 ~\text{GB}$of GPU memory, making their use computationally expensive for hospitals. Therefore, based on these findings, this review highlighted several important areas for continued research development including multicenter models, standardization of image preprocessing and more computationally efficient models.
Institution: Savitribai Phule Pune University
Language: English
Topic: AI in cancer detection
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7165656458