PENGWIN 2026 Task 1 and Task 2: Peripelvic Fracture Segmentation Dataset

The Automatic Segmentation task (Task 1) and Interactive Segmentation task (Task 2) of the PENGWIN 2026 challenge are designed to advance the development of automated peripelvic fracture segmentation methods. The focus is on automatically extracting fracture fragments from 3D CT scans, and utilizing user-provided click prompts to guide models toward high-precision interactive peripelvic segmentation. This repository contains two core dataset modules: Clinical Peripelvic Fracture CT Dataset: This dataset comprises 500 anonymized clinical cases of real peripelvic fractures, covering a wide spectrum of injury patterns including pelvic ring disruptions, acetabular fractures, and proximal femoral fractures (340 for the training set, 160 for the test set). The data is pooled from multiple institutions and imaging systems to ensure high diversity. Each case includes a volumetric 3D CT scan (image.mha) and its corresponding ground-truth peripelvic fracture segmentation mask (label.mha). To distinguish between different bones and fragments, labels are organized into specific integer ranges by bone type: 0: Background 1–50: Sacrum 51–100: Left Hip 101–150: Right Hip 151–200: Femur Each connected fracture fragment is assigned a unique integer value within its respective bone's range. Interactive Click Prompts Dataset: Designed specifically for the Task 2 interactive peripelvic segmentation task, this dataset provides pre-generated click point annotations for each subject in JSON format. To simulate diverse user interaction habits, four distinct click-generation strategies are provided: boundary internal margin (boundary_internal_margin), center of mass (center_of_mass), Euclidean distance transform (euclidean_distance_transform), and uniform random sampling (uniformly_sampled). Click coordinates are provided as voxel indices [x, y, z] in image space, with each click corresponding to exactly one peripelvic fracture fragment. For more information, please visit the PENGWIN2026 challenge webpage. For the PENGWIN 2026 Task 3 dataset (Peripelvic Fracture Reduction Planning), please visit the separate repository (10.5281/zenodo.19734383).