Syn3DWound: A Synthetic Dataset for 3D Wound Bed Analysis

TL;DR

This paper introduces Syn3DWound, an open-source dataset of high-fidelity simulated wounds with 2D and 3D annotations, and proposes baseline methods and a benchmarking framework for automated 3D morphometry analysis and 2D/3D wound segmentation.

Abstract

Wound management poses a significant challenge, particularly for bedridden patients and the elderly. Accurate diagnostic and healing monitoring can significantly benefit from modern image analysis, providing accurate and precise measurements of wounds. Despite several existing techniques, the shortage of expansive and diverse training datasets remains a significant obstacle to constructing machine learning-based frameworks. This paper introduces Syn3DWound, an open-source dataset of high-fidelity simulated wounds with 2D and 3D annotations. We propose baseline methods and a benchmarking framework for automated 3D morphometry analysis and 2D/3D wound segmentation.

Figures (7)

• Figure 1: Syn3DWound aims to produce high-quality synthetic data with precise control of the environment and acquisition protocol from a 2D real-world wound and a 3D avatar. It allows the generation of extensive datasets for evaluating segmentation models. Furthermore, the camera's intrinsic and extrinsic are saved to analyze the performance of 3D reconstruction methods.
• Figure 2: Representations of the specific components involved in the synthetic wound generation. i) 3D human avatar, ii) Wound image and wound extraction, iii) Mesh sculpting including wound shape and placement in the human body, iv) View selection of the 3D human body avatar, and v) Rendering and postprocessing.
• Figure 3: Representations of 2D wound images and their corresponding segmentation maps are generated from various camera trajectories of the 3D wound models. The two models featured in this manuscript are a leg wound (left) and a shoulder wound (right).
• Figure 4: Overview of a traditional framework for 3D reconstruction and analysis: sequential image collection, feature extraction and matching, camera models and sparse point cloud, dense point cloud, meshing point cloud, and 3D reconstruction.
• Figure 5: Evaluating 3D reconstruction outcomes across diverse image resolutions and quantities on the shoulder wound. We showcase results for the COLMAP pipeline using the ASD metric. However, similar trends are observed across different 3D reconstruction methodologies and evaluation metrics.