Refined Machine Learning Model for Medical Image Segmentation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing machine learning models for medical image segmentation and classification face challenges in accurately identifying diverse medical features due to their reliance on general networks trained with large databases, which lack accuracy for underrepresented sub-tasks.

Innovation Solution

A computer-implemented method that involves providing an initial machine learning model, extracting a subset of medical images based on similarity scores, and refining the model using these images to create a specialized model capable of accurately segmenting or classifying specific medical features, even with limited training data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a general machine learning model is trained on a large database of medical images, then the model can handle a wide variety of medical features, but the accuracy for underrepresented sub-tasks deteriorates

Engineering Contradiction:

Improvecapability to handle diverse medical featuresVSAvoidaccuracy for underrepresented sub-tasks

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent segments the training process into two distinct phases: first training a general model on the entire diverse database to learn universal features, then creating task-specific refined models by retraining on subsets of images relevant to specific underrepresented sub-tasks. This segmentation allows the system to maintain both general versatility and specific accuracy.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by creating specialized refined models with enhanced capabilities for specific sub-tasks while maintaining the general model for broader applications. Each refined model is optimized for particular medical features or conditions, ensuring high accuracy for underrepresented cases without sacrificing the ability to handle diverse medical features overall.

Inventive Principle:

Principle #3Local quality

2Measurement precision

If a machine learning model is specialized for a specific medical feature type, then accuracy for that feature improves, but the ability to handle diverse medical features deteriorates

Engineering Contradiction:

Improveaccuracy for specific medical featureVSAvoidcapability to handle diverse medical features

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent implements a dynamic system where the appropriate model (general or refined) is selected based on the specific task at hand. The system can switch between the general model for diverse features and specialized refined models for specific underrepresented sub-tasks, optimizing performance adaptively rather than being fixed in one configuration.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent creates a multi-functional framework where a single system can perform both general medical image analysis using the base model and specialized analysis using refined models. The system maintains multiple models with different specializations, allowing it to universally handle various medical features while providing expert-level accuracy for specific conditions when needed.

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Measurement precision

If training data for a specific medical condition is limited, then the model cannot learn specific features adequately, but using more diverse data dilutes the specialization

Engineering Contradiction:

Improveaccuracy for specific conditionVSAvoidamount of training data for specific condition

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent performs preliminary action by first training the general model on the complete diverse database to establish a strong foundation of universal medical image features. This preliminary training ensures that even when training data for a specific condition is limited, the refined model inherits robust general knowledge that compensates for the limited specialized data.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The general model serves as an intermediary that bridges the gap between limited specific condition data and the need for accurate specialization. By refining the general model with the available specific condition data, the system leverages the intermediary's pre-learned features to achieve high accuracy even with small training sets for underrepresented conditions.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS11334992B2Computer-implemented method to increase the accuracy of machine learning models for medical image processing

Publication Date: 2022.05.17 KONINKLIJKE PHILIPS NV

AI summary

There is provided a computer implemented method (200) for medical image processing. The method comprises providing (202) a database of medical images and providing (204) an initial machine learning model which is trained for segmenting or classifying a medical feature in the medical images. The method also comprises extracting (206) a subset of medical images from the database based on a similarity score of the medical images and training (208) the machine learning model using the extracted subset of medical images in order to provide a refined machine learning model.