An Exceptional Dataset For Rare Pancreatic Tumor Segmentation

TL;DR

This work tackles the scarcity of annotated imaging data for pancreatic neuroendocrine tumors (pNETs) and the consequent segmentation challenges on CT scans. It introduces the first dedicated pNET CE-CT dataset, comprising 469 patient volumes with multi-phase arterial and venous data and rigorous slice-level ground-truth masks, along with a pancreas-focused post-processed version. The authors evaluate several UNet-based baselines and propose a slice-wise weighted Dice loss to mitigate extreme class imbalance and improve detection of small lesions. By providing a specialized resource and demonstrating improved segmentation with targeted loss weighting, the paper aims to accelerate the development of accurate pNET diagnostics and improve preoperative assessment and patient outcomes.

Abstract

Pancreatic NEuroendocrine Tumors (pNETs) are very rare endocrine neoplasms that account for less than 5% of all pancreatic malignancies, with an incidence of only 1-1.5 cases per 100,000. Early detection of pNETs is critical for improving patient survival, but the rarity of pNETs makes segmenting them from CT a very challenging problem. So far, there has not been a dataset specifically for pNETs available to researchers. To address this issue, we propose a pNETs dataset, a well-annotated Contrast-Enhanced Computed Tomography (CECT) dataset focused exclusively on Pancreatic Neuroendocrine Tumors, containing data from 469 patients. This is the first dataset solely dedicated to pNETs, distinguishing it from previous collections. Additionally, we provide the baseline detection networks with a new slice-wise weight loss function designed for the UNet-based model, improving the overall pNET segmentation performance. We hope that our dataset can enhance the understanding and diagnosis of pNET Tumors within the medical community, facilitate the development of more accurate diagnostic tools, and ultimately improve patient outcomes and advance the field of oncology.

Figures (6)

• Figure 1: Demonstration of pancreas in human bodyACS2024. Pancreas consists both endocrine cells and exocrine cells. Most pancreatic cancers are characterised as ductal adenocarcinoma and thus represent malignancy of the exocrine pancreas whereas a minority represent neuroendocrine tumoursmizrahi2020pancreatic. Our work focus on this less populated type of pancreatic cancer.
• Figure 2: Examples of our pNETs dataset. Each row represents a patient's CECT image. (a), (b), (c) illustrates the image and annotated tumor area in Arterial phase, (d) illustrates the Venous phase.
• Figure 3: Histogram of tumor area for the sliced 2-D images containing tumor in our dataset. Y-axis counts the amount of pixels marked positive in the GT mask. We see that most of the images are only with a small tumor mask, resulting in a skewed histogram.
• Figure 4: The framework of our proposed baseline method. For each 2D CT image, we first segment out the pancreas mask from the given CT image using pretrained models. Next, we crop and augment the area and send the corresponding image into our model for pNET segmentation.
• Figure 5: Our proposed slice-wise weight loss, which assigns sampling weights to each 2D CT image based on the relative ranking of its tumor area among all slices from the same patient.