Surgery Scene Restoration for Robot Assisted Minimally Invasive Surgery

TL;DR

This paper introduces a novel online preprocessing framework that restores clean, sharp images from noisy, blurred, and raw observations in robot-assisted minimally invasive surgery, improving visualization and aiding surgical tasks.

Contribution

It proposes a new method for real-time image restoration in MIS, addressing noise, blur, and color correction challenges caused by hardware and environmental factors.

Findings

Effective noise and blur removal in surgical images

Improved visualization for surgical scene understanding

Compatible with real-time processing constraints

Abstract

Minimally invasive surgery (MIS) offers several advantages including minimum tissue injury and blood loss, and quick recovery time, however, it imposes some limitations on surgeons ability. Among others such as lack of tactile or haptic feedback, poor visualization of the surgical site is one of the most acknowledged factors that exhibits several surgical drawbacks including unintentional tissue damage. To the context of robot assisted surgery, lack of frame contextual details makes vision task challenging when it comes to tracking tissue and tools, segmenting scene, and estimating pose and depth. In MIS the acquired frames are compromised by different noises and get blurred caused by motions from different sources. Moreover, when underwater environment is considered for instance knee arthroscopy, mostly visible noises and blur effects are originated from the environment, poor control on illuminations and imaging conditions. Additionally, in MIS, procedure like automatic white balancing and transformation between the raw color information to its standard RGB color space are often absent due to the hardware miniaturization. There is a high demand of an online preprocessing framework that can circumvent these drawbacks. Our proposed method is able to restore a latent clean and sharp image in standard RGB color space from its noisy, blur and raw observation in a single preprocessing stage.