Remi Chierchia,Leo Lebrat,David Ahmedt-Aristizabal,Olivier Salvado,Clinton Fookes,Rodrigo Santa Cruz
Abstract:Managing chronic wounds is a global challenge that can be alleviated by the adoption of automatic systems for clinical wound assessment from consumer-grade videos. While 2D image analysis approaches are insufficient for handling the 3D features of wounds, existing approaches utilizing 3D reconstruction methods have not been thoroughly evaluated. To address this gap, this paper presents a comprehensive study on 3D wound reconstruction from consumer-grade videos. Specifically, we introduce the SALVE dataset, comprising video recordings of realistic wound phantoms captured with different cameras. Using this dataset, we assess the accuracy and precision of state-of-the-art methods for 3D reconstruction, ranging from traditional photogrammetry pipelines to advanced neural rendering approaches. In our experiments, we observe that photogrammetry approaches do not provide smooth surfaces suitable for precise clinical measurements of wounds. Neural rendering approaches show promise in addressing this issue, advancing the use of this technology in wound care practices.
What problem does this paper attempt to address?
The primary aim of this paper is to address the challenges in chronic wound management, particularly by alleviating this global health and economic burden through clinical wound assessment from consumer-grade videos using automated systems. The core contributions of the paper include:
- **Proposing the SALVE Dataset**: To evaluate the effectiveness of 3D wound reconstruction techniques, the authors introduce a new dataset called SALVE, which includes video recordings of realistic wound simulants (silicone models) captured using different cameras. These data aim to capture challenges common in clinical settings, such as complex lighting conditions and variations in image quality.
- **Comprehensive Study on 3D Wound Reconstruction**: Utilizing the SALVE dataset, the authors assess the accuracy and precision of 3D wound reconstruction from consumer-grade videos, covering the latest techniques from traditional photogrammetry pipelines to advanced neural rendering methods. The study particularly focuses on ensuring that measurements obtained from the reconstructed geometry are independent of the viewpoint and can calculate richer wound biomarkers.
- **Benchmarking Multiple Methods**: The paper compares various 3D reconstruction methods, including traditional photogrammetry tools (such as COLMAP), modern neural rendering methods (such as Neuralangelo and Neus-facto), and methods based on 3D Gaussian Spheres and 2D Gaussian Spheres (SuGaR and 2DGS). By comparing the performance of these methods on the SALVE dataset, it is found that neural rendering methods have an advantage in handling complex wound geometries, generating smooth and accurate surfaces suitable for clinical wound measurement.
- **Evaluation Metrics**: To quantify the performance of different methods, the paper defines a series of evaluation metrics, including Average Distance (AD), Earth Mover's Distance (W2), Hausdorff Distance (HD and its 90th percentile version HD 90), Normal Consistency (NC), and Normal Consistency under Optimal Transport Coupling (W2-NC). These metrics help evaluate the accuracy of the reconstruction and the smoothness of the reconstructed surfaces.
- **Experimental Results**: According to the results provided in the paper, the methods Meshroom, Neuralangelo, and Neus-facto perform best across multiple evaluation criteria. Notably, Neus-facto excels in handling complex wound geometries and shows strong robustness to variations in input image quality. Additionally, Neus-facto performs well in consistency evaluations across different devices, indicating its high reliability in practical applications.
In summary, this paper, by proposing a comprehensive benchmarking framework, not only provides valuable resources for research in the field of 3D wound reconstruction but also promotes the development of this field, particularly in improving the accuracy and practicality of automated wound assessment systems.