Medical Data Model Selection for Reliable Condition Indication

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current medical data processing techniques face challenges such as poor generalization, low robustness, and lack of explainability, particularly when dealing with different data modalities and qualities, leading to unreliable and untrustworthy medical condition determinations.

Innovation Solution

A method that dynamically selects suitable machine learning models based on specific properties of medical data, using an ensemble of models to provide a reliable indication of a medical condition by aggregating model-specific indications, prioritizing models with high empirical performance and explainability, and utilizing unlabeled training data for anomaly detection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a single machine learning model is used for medical condition determination, then the device complexity is low, but the reliability and robustness of the determination deteriorate when dealing with different data modalities and qualities

Engineering Contradiction:
Improvereliability of medical condition determinationVSAvoidcomplexity of model selection system
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The system dynamically selects models from the ensemble based on properties of the input medical data (modality, quality, resolution) rather than using a fixed single model. This dynamic adaptation allows the system to optimize reliability for each specific data type while maintaining manageable complexity through automated selection rules.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The ensemble of models is designed to handle multiple types of medical data (CT images, MRI images, X-rays, EEG signals) with different qualities and resolutions. Each model in the ensemble is trained on specific data types, creating a universal system that can reliably process diverse medical data modalities through appropriate model selection.

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Measurement precision

If deep learning models are used to improve diagnostic quality, then the measurement precision is improved, but the explainability and traceability of results deteriorate due to black box nature

Engineering Contradiction:
Improvediagnostic qualityVSAvoidexplainability of prediction
Core Design Contradiction:
Measurement precisionVSLoss of information

Solution Approach 1:

The patent introduces model selection rules and aggregation mechanisms as intermediaries between the input data and final prediction. These intermediaries provide traceability by documenting which models were selected based on data properties and how their predictions were aggregated, making the black box deep learning models more explainable without reducing their diagnostic precision.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The system segments the prediction process into distinct stages: data property analysis, model selection, individual model prediction, and result aggregation. This segmentation allows each stage to be independently analyzed and explained, improving overall system explainability while maintaining the high precision of deep learning models.

Inventive Principle:
Principle #1Segmentation

3Adaptability or versatility

If models are trained on specific data modalities to improve generalization, then the adaptability to different data types is improved, but the device complexity increases due to needing multiple specialized models

Engineering Contradiction:
Improveadaptability to different data modalitiesVSAvoidcomplexity of ensemble model system
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system performs preliminary analysis of input data properties (modality, quality, resolution) before model selection. This preliminary action allows the system to automatically match the appropriate pre-trained specialized model to the input data type, achieving high adaptability across different medical data modalities while keeping the system complexity manageable through automated matching rules.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system changes the parameter of model selection based on input data properties. By dynamically adjusting which model from the ensemble is activated based on data modality, quality, and resolution parameters, the system achieves versatile adaptability to different medical data types without requiring a single complex universal model.

Inventive Principle:
Principle #35Parameter changes

4Reliability

If multiple models are used to improve robustness, then the reliability of determination is improved, but the loss of time for processing increases due to aggregating multiple model outputs

Engineering Contradiction:
Improverobustness of medical condition determinationVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system dynamically determines how many models from the ensemble to activate based on input data properties and confidence requirements. For routine cases, fewer models may be sufficient, reducing processing time. For uncertain or critical cases, more models are activated to maximize reliability, optimizing the trade-off between robustness and processing time.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system applies partial action by selectively activating only the necessary subset of models from the ensemble based on data properties rather than always running all models. This approach maintains high reliability by using multiple specialized models when needed while reducing processing time by avoiding unnecessary model executions for straightforward cases.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3872584A1Technique for determining an indication of a medical condition
Publication Date: 2021.09.01 DEEPC GMBH
  • EP3872584A1 patent drawingFigure 1
  • EP3872584A1 patent drawingFigure 2
  • EP3872584A1 patent drawingFigure 3

AI summary

A medical data processing technique for determining an indication of a medical condition is disclosed. A method implementation of the technique comprises selecting (202), based on at least one property associated with medical data of a test instance, at least one model out of a plurality of models, wherein each of the plurality of models is generated by a learning algorithm and configured to provide a model-specific indication of the medical condition, determining (204), using each of the at least one selected model, a respective model-specific indication, and determining (206), based on the model-specific indications, the indication of the medical condition.