Bayesian Network Diagnostic Model Integrating Sensitivity Specificity

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for diagnosing diseases using Bayesian networks often lead to errors due to the dismissal of potential diseases or consideration of too broad a base of possibilities, as they do not adequately account for the specificity and sensitivity of signs, which can vary significantly between individuals and clinical forms of diseases.

Innovation Solution

A system and method that integrates sensitivity and specificity statistics into Bayesian network modeling, allowing for the acquisition and recording of patient-specific factors, signs, and disease data to generate more reliable lists of conditional probabilities associated with diseases, while also considering the influence of medical products and their active ingredients.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If logical influence links are used to model disease-diagnosis relationships, then the model structure is simple, but the diagnostic accuracy deteriorates due to inability to account for sensitivity and specificity variations

Engineering Contradiction:
Improvemodel structureVSAvoiddiagnostic accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent transforms the binary logical link model into a probabilistic model by introducing sensitivity and specificity parameters. Each sign-disease relationship is characterized by these parameters, allowing the model to account for variations in diagnostic reliability across different signs and diseases, thereby improving diagnostic accuracy while maintaining model tractability

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent adds a new dimension to the model by incorporating sensitivity and specificity statistics as additional parameters. This transforms the simple presence/absence logical links into a multi-dimensional probabilistic framework that captures the nuanced relationships between signs, diseases, and diagnostic reliability

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Measurement precision

If sensitivity and specificity statistics are integrated into Bayesian network modeling, then diagnostic accuracy is improved, but computational complexity increases

Engineering Contradiction:
Improvediagnostic accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent performs preliminary calculations of sensitivity and specificity statistics during the model construction phase. These pre-computed values are then used directly in the Bayesian network calculations, avoiding the need for complex real-time computations during diagnosis and reducing overall computational burden

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system automatically computes and stores sensitivity and specificity statistics from medical literature and clinical data. These pre-computed values serve the diagnostic engine directly, eliminating the need for manual parameter specification and reducing computational complexity during actual diagnosis operations

Inventive Principle:
Principle #25Self-service

3Adaptability or versatility

If a broad base of potential diseases is considered, then the model covers more possibilities, but the reliability of individual probability estimates deteriorates

Engineering Contradiction:
Improvedisease coverageVSAvoidprobability estimate reliability
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent applies different levels of probabilistic modeling to different disease-sign relationships based on their specific characteristics. Signs with well-established sensitivity and specificity values receive more precise probabilistic treatment, while less certain relationships use broader ranges or expert elicited values, allowing the model to maintain high reliability across diverse disease possibilities

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP3821445B1System and method for generating a list of probabilities associated with a list of diseases, computer program product
Publication Date: 2024.08.07 E DIAGMED
  • EP3821445B1 patent drawingFigure 1
  • EP3821445B1 patent drawingFigure 2
  • EP3821445B1 patent drawingFigure 3

AI summary

A method for generating a list of probabilities (LIST1) associated with a list of diseases for a first patient (P1), said method comprising: ▪ First acquisition (ACQ1) of a first set of patient data (ENS1) of the first patient (P1), comprising: • an age value (AGE1), • a gender value (GEN1); ▪ Second acquisition (ACQ2) of data describing at least one disease (ANT1) of said patient (P1), said disease (ANT1) being extracted from a first database (BDM), each disease (ANT1) being associated with a first prevalence statistic (PR1) and a first incidence statistic (IN1), and each disease being associated with a list of signs; ▪ Third acquisition (ACQ3) of data describing at least one first sign (S1), said first sign (S1) comprising a first sensitivity statistic (SEi) and a second, specificity statistic (SPi) for each disease of a predefined list (LIST1) of diseases (ANT1) associated with said sign; ▪ Generation (GEN), from a first model of a Bayesian network (RB) and input data comprising the data from the first, second and third acquisitions (ACQ1, ACQ2, ACQ3), of a set of probabilities, each probability being associated with a given disease from the first list (LIST1).