semi-synthetic training datasets
收藏资源简介:
本研究开发了半合成训练数据集,结合模拟数据和体内测量数据,以解决真实数据获取困难和纯模拟数据真实性不足的问题。数据集包含2000张图像,用于训练深度学习模型,以增强LED基光声成像中的临床金属针头可见性。该模型在血-血管模拟幻影、离体猪肉组织和人体手指的体内成像中进行了评估,显著提高了针头在光声成像中的可见性,有助于减少经皮针插入过程中的并发症。
This study developed a semi-synthetic training dataset that integrates simulated data and in vivo measured data, to resolve the dual issues of limited availability of real-world clinical data and the poor authenticity of purely simulated datasets. The dataset consists of 2000 images, which are utilized to train deep learning models for improving the visibility of clinical metallic needles in LED-based photoacoustic imaging. The trained model was evaluated via in vivo imaging of blood-vessel phantoms, ex vivo porcine tissue, and human fingers, and the experiments demonstrated that the model significantly enhanced the visibility of needles in photoacoustic imaging, thereby helping to reduce complications associated with percutaneous needle insertion procedures.

- 1Improving needle visibility in LED-based photoacoustic imaging using deep learning with semi-synthetic datasets伦敦国王学院生物医学工程与成像科学学院 · 2022年



