Research on endoscopic blood scattering medium imaging method based on spatial similarity characteristics of light field and deep learning

With the rising incidence of spinal disorders, demand for minimally invasive spinal surgery using endoscopes continues to grow, owing to advantages such as reduced blood loss, minimal trauma, and rapid recovery. As the core intraoperative imaging device, the clarity of endoscopic imaging directly impacts surgical outcomes. However, as a form of underwater minimally invasive surgery, spinal endoscopy frequently involves intraoperative bleeding. The strong absorption and multiple scattering caused by blood flow constitute a dynamic scattering medium interference, severely compromising imaging quality during surgery. This leads to difficulties in locating bleeding points and prolonged operative times. To mitigate the impact of intraoperative blood scattering media on endoscopic imaging quality, this paper proposes a spinal endoscopic blood scattering medium imaging method based on spatial similarity cross-correlation of light fields and deep learning. This method first performs fusion preprocessing on the multi-angle raw scattered subaperture images acquired via light field capture. Subsequently, the study conducts full-scale structural analysis on the blurred images captured by the light field camera. Building upon the existing U-Net network principle, the network architecture is enhanced to construct a full-scale U-Net network, which is employed to restore the pre-processed images. Ultimately, experiments validate the method's efficacy, using this method to restore images of blood scattering media transmitted through multiple turbidity levels resulted in an average improvement of 76.95% in the image SSIM metric, representing a significant advancement over existing approaches. The proposed method offers valuable insights for subsequent research into blood-scattering medium imaging techniques within complex and realistic endoscopic surgical scenarios. It holds significant clinical application value in facilitating surgical procedures and mitigating surgical risks.