Spectrogram Biomarker Grouping for Accurate Non-Invasive Blood Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current blood biomarker testing methods require invasive blood draws and laboratory analysis, which are time-consuming and stressful for patients, and existing non-invasive techniques lack accuracy due to overfitting issues.
Innovation Solution
A method for grouping blood biomarkers based on concentration levels and biological correlations, using spectrogram data to train prediction models that predict biomarker levels through multiple routes, enhancing accuracy and reducing invasiveness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional blood draw and laboratory analysis methods are used, then measurement precision of blood biomarkers is achieved, but loss of time increases and ease of operation deteriorates
Solution Approach 1:
The patent replaces the mechanical/chemical laboratory analysis system with an optical detection system. Spectrometric sensors detect blood biomarkers through non-invasive optical measurements, substituting the traditional mechanical blood draw and chemical laboratory testing process. This enables rapid results without laboratory infrastructure while maintaining measurement capability.
Solution Approach 2:
The patent creates a computational model that copies the relationship between spectrometric data and blood biomarker concentrations. By training prediction models on paired spectrometric measurements and laboratory-confirmed biomarker levels, the system replicates laboratory accuracy through software algorithms rather than physical chemical analysis.
2Ease of operation
If non-invasive spectrometric methods are used, then ease of operation improves and loss of time decreases, but measurement precision deteriorates due to overfitting
Solution Approach 1:
The patent segments the prediction model into multiple independent models, each specialized for predicting specific blood biomarker types (e.g., glucose, cholesterol, hemoglobin). This segmentation prevents overfitting by limiting each model's scope to relevant biomarker characteristics, improving generalization accuracy while maintaining non-invasive convenience.
Solution Approach 2:
The patent implements dynamic model selection that adapts to different measurement conditions, patient populations, and biomarker types. The system dynamically adjusts prediction strategies based on input data characteristics, preventing static model overfitting while maintaining ease of non-invasive operation across diverse scenarios.
3Measurement precision
If multiple prediction models are used to improve accuracy, then device complexity increases
Solution Approach 1:
The patent creates a universal prediction framework where multiple specialized models operate under a single integrated system. The common data processing pipeline, feature extraction methods, and model selection logic serve all biomarker predictions, reducing overall system complexity despite using multiple prediction models for different biomarker types.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The method provides rapid, non-invasive blood biomarker analysis with improved accuracy by grouping biomarkers and using multiple prediction routes, overcoming overfitting and reducing the need for laboratory testing.
Implementation Method 1
spectrometric measurements
Implementation Method 2
spectrometric reading from a patient's tissue
Data Source
AI summary
A system, method and corresponding software product are presented, the method comprising: providing a training data set comprising one or more spectrogram data pieces obtained from a plurality of individuals and respective data on a selected set of blood biomarkers of said individuals; selecting one or more groups of biomarkers selected from said selected set of biomarkers, wherein each group includes two or more (three or more) biomarkers; training one or more prediction models based on said training data, said one or more prediction model comprising one or more prediction routes for prediction of said one or more groups of biomarkers respectively. Accordingly, the prediction model comprises a selected number of prediction routes, each trained for predicting biomarkers concentrations of a respective groups of biomarkers. The biomarkers may be selected into groups in accordance with biological or biochemical correlations between them, or in accordance with concentration levels of the biomarkers.