Surgical Image Feature Compression for Edge Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robotic surgical systems face challenges in processing large amounts of image data due to high computational resource requirements, label annotation costs, privacy concerns, and inconsistent model performance across different surgical maneuvers, leading to increased latency and resource consumption.
Innovation Solution
Implementing a self-supervised machine learning model to extract features from surgical images, allowing for compressed representations that can be processed at edge devices, enabling efficient downstream analysis and improved anomaly detection while maintaining data privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution imaging is used to capture surgical procedures, then image quality and detail are improved, but computing, memory and networking resources are significantly consumed
Solution Approach 1:
The patent extracts only the essential features from high-resolution surgical images using machine learning models, rather than processing or storing the complete high-resolution images. This extraction approach captures the critical diagnostic information while discarding redundant data, thereby maintaining measurement precision for surgical analysis while dramatically reducing computing, memory and networking resource consumption.
Solution Approach 2:
The patent segments the image processing task into two stages: first, extracting key features from high-resolution images using trained models; second, processing only these extracted features for storage and analysis. This segmentation allows the system to benefit from high-resolution imaging quality while avoiding the resource burden of processing complete high-resolution images throughout the workflow.
2Loss of information
If complete high-resolution images are stored for later review, then data availability for analysis is improved, but disk space consumption increases significantly
Solution Approach 1:
The patent extracts essential diagnostic features from surgical images and stores only these extracted features rather than the complete images. This extraction maintains data availability for later review and analysis while dramatically reducing disk space requirements, as the feature representations are much more compact than full-resolution images.
Solution Approach 2:
The patent creates compressed representations (copies) of the essential image information in the form of extracted features. These feature copies retain the critical diagnostic content needed for surgical review and analysis while occupying minimal storage space compared to storing original high-resolution images.
3Use of energy by moving object
If features are extracted using machine learning models, then resource consumption is reduced, but model training requires significant computational resources and labeled data
Solution Approach 1:
The patent employs self-supervised learning approaches where the machine learning models learn to extract meaningful surgical features without requiring extensive manual annotation of training data. The models are trained to recognize patterns and features that are inherently present in the surgical images, allowing them to serve themselves by learning from the raw image data without heavy reliance on externally provided labeled datasets, thereby reducing the complexity and cost of model training.
4Power
If surgical images are processed centrally in datacenters, then processing power is sufficient, but latency increases and edge devices cannot process images locally
Solution Approach 1:
The patent extracts essential features from surgical images that can be processed with minimal computational resources. These extracted features are compact and contain the critical information needed for surgical analysis, enabling them to be processed efficiently on edge devices with limited processing power, thereby reducing latency and eliminating the need for centralized datacenter processing.
Solution Approach 2:
The patent enables a dynamic processing architecture where the level of processing can adapt to available resources. The extracted features serve as an intermediate representation that can be processed either locally on edge devices for low-latency applications or transmitted to centralized systems when higher processing power is available, providing flexibility in processing location and timing.
Data Source
AI summary
Extracting features to compress images generated during medical procedures is provided. In examples, systems are configured to obtain one or more frames of a video captured by a camera of a medical procedure performed with a robotic medical system. Systems can be configured to generate features for the one or more frames using a first model trained with self-supervised machine learning and constructing a dataset based on the generated features. Some systems can be configured to construct a dataset based on the generated features and input the dataset into a second model to detect an aspect of the medical procedure.