Jiun-In Guo, PhD1, Kai-Che Liu, PhD2, Chia-Chi Tsai1, Jan-Yang Lin1, Yong-Cyuan Yang1, Ching-Hwa Cheng, PhD3, Jungle Chi-Hsiang Wu, Dr2, Shih-Wei Huang, Dr4. 1National Chiao Tung University, 2Chang Bing Show Chwan Memorial Hospital, 3Feng Chia University, 4Show Chwan Memorial Hospital
Introduction: The proposed image processing method aims to develop the real-time image processing method for endoscope defogging. Minimally invasive surgery (MIS) is the current surgical trend because MIS can provide a lot of advantages to both patients and surgeons, including less bleeding, less scar and higher surgical quality. Due the development of surgical technology, MIS uses smaller and smaller diameter endoscope, for example, 4mm for laparoscope. Image quality of endoscope is a very import issues. In current surgical situations, endoscope fogging sometimes happens and makes the surgery not smoothly. There are several methods used currently to avoid fogging, such as applying anti-fogging agent or heating, however, those procedures must be frequently applied during the surgery. On the other hand, some image-based processing was proposed to defog. Since several images are needed for those image-based processing methods, those algorithms are not suitable to apply in MIS. Therefore, the proposed image processing defogging method is based on single image calculation and is real-time.
Methods and Procedures: The proposed image processing method includes several procedures, starting from exposure control, harsh environment judgment, histogram redistribution, and new pixel interpolation by probability transfer function. The captured endoscope image is divided into several segments to calculate the average brightness. Harsh environment judgment is taken image contrast value after exposure control of brightness to separate fogging area. Image histogram will be redistributed to equalize fogging area pixel value into the average value so that the fogging situation will be removed. Then the pixel interpolation will be performed to fill the removed area. A transfer function is used based on the probability distribution of each segment. In order to achieve real-time operation, the algorithm is implemented on parallel processing.
Results: The proposed image processing method is implemented on Nvidia® Tegra X1 platform which contains Quad-core ARM® Cortex®-A57 MPCore Processor and 256 CUDA cores for parallel processing to achieve Full HD 30Hz image output. Figure 1 shows the defogging results based on the proposed image processing method where (a)(c)(e)(g) are images with fogging and (b)(d)(f)(h) are defogging results accordingly based on the proposed method. A clear image is obtained without the fogging.
Conclusions: The proposed image processing method is based on single image and can be real-time implemented for endoscope defogging during minimally invasive surgery. Parallel processing architecture is applied to achieve Full HD 30Hz performance. Defogging results show the possibility in real clinical applications.
Presented at the SAGES 2017 Annual Meeting in Houston, TX.
Abstract ID: 79532
Program Number: P413
Presentation Session: Poster (Non CME)
Presentation Type: Poster
