Laparoscopic Smoke Removal via GAN Mask Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for laparoscopic image smoke removal, such as those based on atmospheric scattering models, dark channel priori models, and Bayesian inference models, are inadequate due to their limitations in predicting smoke distribution and maintaining image details, especially in the complex and non-uniform scenarios of laparoscopic surgery, leading to unsatisfactory results and increased treatment costs.
Innovation Solution
A laparoscopic image smoke removal method using a generative adversarial network (GAN) that includes a smoke mask segmentation network and a multi-level smoke feature extractor, which processes images to generate smoke-free images by filtering out smoke information while maintaining real details, employing a dual-domain generator and discriminator network for adversarial training and cycle consistency loss to improve texture details.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If atmospheric scattering model is used for smoke removal, then global atmospheric light can be estimated, but the method fails when smoke concentration varies greatly and light source is close to tissues
Solution Approach 1:
The patent transitions from global atmospheric light estimation to local transmission map estimation, changing the spatial scale parameter. It also shifts from assuming parallel light rays to modeling divergent light paths from close light sources, fundamentally changing the geometric parameters of the model to suit laparoscopic conditions.
Solution Approach 2:
The patent applies local quality by estimating transmission maps locally rather than globally. The method processes different regions of the image with locally adapted parameters, allowing the smoke removal to be effective in areas with varying smoke concentrations and different lighting conditions throughout the surgical field.
2Measurement precision
If dark channel priori model is used for smoke removal, then pixels with low intensity values can be identified, but specular reflection causes problems due to large color difference and short light source distance
Solution Approach 1:
The patent extracts and removes the harmful specular reflection components from the image processing pipeline. By using a learned transmission map estimation approach rather than relying on dark channel assumptions, it separates the smoke removal function from the problematic dark channel constraint, effectively taking out the source of reflection interference.
3Measurement precision
If Bayesian inference model is used for smoke removal, then color and texture of smoke images can be modeled, but image distortion occurs due to over-enhancement
Solution Approach 1:
The patent implements feedback through iterative optimization of the transmission map estimation. The method uses the estimated transmission map to guide smoke removal while continuously refining the estimation based on the resulting image quality, preventing over-enhancement and maintaining natural appearance of smoke-free regions.
4Object-affected harmful factors
If mechanical smoke removal device is used, then surgical smoke can be physically removed, but operation duration is prolonged and treatment cost increases
Solution Approach 1:
The patent replaces the mechanical smoke removal system with a computational image processing system. Instead of using physical devices to remove smoke from the surgical field, it uses algorithms to remove smoke artifacts from the captured images, eliminating the need for mechanical intervention and associated time losses.
Solution Approach 2:
The patent creates a virtual copy of the smoke-free surgical field through image processing. Rather than physically removing smoke, it generates a cleaned version of the smoke-containing image, providing surgeons with clear visual information without altering the actual surgical environment or extending operation time.
5Measurement precision
If existing deep learning smoke removal models are used, then smoke can be removed from images, but network model cannot be embedded in laparoscopic equipment for real-time use
Solution Approach 1:
The patent segments the deep learning model into modular components that can be efficiently deployed on embedded systems. By dividing the complex smoke removal task into discrete processing stages with optimized computational requirements, the model becomes suitable for integration into laparoscopic equipment with limited processing resources.
Data Source
AI summary
A laparoscopic image smoke removal method based on a generative adversarial network, and belongs to the technical field of computer vision. The method includes: processing a laparoscopic image sample to be processed using a smoke mask segmentation network to acquire a smoke mask image; inputting the laparoscopic image sample to be processed and the smoke mask image into a smoke removal network, and extracting features of the laparoscopic image sample to be processed using a multi-level smoke feature extractor to acquire a light smoke feature vector and a heavy smoke feature vector; and acquiring, according to the light smoke feature vector, the heavy smoke feature vector and the smoke mask image, a smoke-free laparoscopic image by filtering out smoke information and maintaining a laparoscopic image by using a mask shielding effect. The method has the technical effects of robustness and ability of being embedded into a laparoscopic device for use.