Abstract: In order to address the problems of slow generation and poor image quality of digital reconstructed radiographs (DRR) in the alignment of preoperative 3D computed tomography (CT) images with intraoperative 2D X-ray images, a single-scale Retinex-based improved grey-scale image enhancement method is proposed. The proposed method is based on a single-scale Retinex improvement of grey-scale image enhancement and employs a CUDA-based GPU parallel light projection algorithm globally. A hierarchical alignment method based on a weighted histogram of gradient directions (WHGD) is proposed to address the problems of slow convergence and small capture range caused by simultaneous optimization of all pose parameters in the 2D/3D alignment algorithm. The WHGD is independent of the translation parameters to achieve spatial decoupling of the parameters, and an adaptive evolutionary optimization strategy of the covariance matrix is introduced to ensure correct convergence of all the positional parameters while avoiding local optima. The experimental results show that the method is 41.42% faster and 34.53% more accurate than similar alignment methods, meeting the requirements for accuracy and speed in clinical settings.

Key words: 2D/3D alignment, similarity metrics, gradient direction histograms, minimally invasive image guided surgery