Medical Image Processing With Neural Distortion and Tissue Screening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems for medical applications, particularly in time-critical surgical scenarios, struggle to accurately and efficiently distinguish relevant medical images from distorted or irrelevant ones, leading to potential misinterpretation and lack of trust among practitioners.
Innovation Solution
A two-step filtering process using trained neural networks to identify and remove image distortions and tissue types, followed by a third neural network for segmentation, allowing practitioners to visualize and analyze only relevant and high-quality images with highlighted segments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple images are provided to the pathologist at high speed during surgery, then the surgical procedure can be supported with real-time image data, but the pathologist has limited time to analyze each image which increases the risk of incorrect evaluation
Solution Approach 1:
The patent segments the image review process into two stages: an automated preliminary screening stage that filters images based on distortion detection, and a manual review stage for pathologists. This segmentation allows high-speed processing of large image datasets while maintaining evaluation accuracy by pre-filtering out distorted images before human review.
Solution Approach 2:
The patent introduces an intermediary automated image processing system that acts as a mediator between the high-speed image acquisition and the pathologist's review. This intermediary system detects and filters distorted images, reducing the burden on pathologists while maintaining review quality.
2Productivity
If automated solutions are used to determine image relevance, then the review process can be accelerated, but the solutions can only inaccurately determine whether an image is relevant and cannot provide understandable decision-making
Solution Approach 1:
The patent implements feedback mechanisms where pathologists can provide annotations and corrections on image relevance and distortion detection. This feedback loop continuously improves the automated system's accuracy in determining image relevance while maintaining explainable decision-making through visual feedback displays.
Solution Approach 2:
The patent replaces manual visual inspection with automated neural network-based image analysis systems that can accurately determine image relevance and distortion. This substitution maintains high productivity while improving measurement precision through advanced machine learning models.
3Extent of automation
If practitioners trust automated image processing solutions, then acceptance and adoption increase, but lack of interaction possibilities and understandable decision-making leads to lack of trust
Solution Approach 1:
The patent creates a dynamic interactive interface that allows pathologists to adjust filtering parameters, review flagged images, and provide feedback on automated decisions. This dynamic interaction improves practitioner acceptance by giving them control over the automated process while maintaining high automation extent.
Solution Approach 2:
The patent implements comprehensive feedback mechanisms including visual explanations of automated decisions, annotation tools for pathologists to correct misclassifications, and continuous improvement of the automated system based on practitioner feedback. This feedback loop builds trust by making the automated system more transparent and adaptable to practitioner needs.
Data Source
AI summary
A method for processing of images includes inputting the plurality of images into a trained first neural network and receiving, as an output of the first neural network and for each of the plurality of images, at least one first classifier indicating a set image distortion type of the image, identifying a first subset of the plurality of images based on the first classifiers, inputting the first subset of images into a trained second neural network different from the first neural network and receiving, as an output of the second neural network and for each of the first subset of images, at least one second classifier indicating a presence of a set tissue type in the image, and identifying a second subset of the first subset of images based on the second classifiers. A processor unit and a system are configured to perform such a method.