Fundus Image Analysis for Relative Cardiovascular Aging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current CVD risk prediction methods using retinal images are inaccurate due to training against single labels (chronological age or conventional equations) and fail to identify major contributors, leading to many false positives and negatives, necessitating more precise CVD risk stratification for targeted treatment.
Innovation Solution
A method using deep learning models to analyze fundus images for determining relative cardiovascular aging (Cardiac BioAge) by comparing an individual's predicted CVD risk with a set of similar-aged individuals, incorporating factors like blood pressure, cholesterol, and glycaemic control, and adjusting for device-specific image processing to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current statistical methods with regression models are used for CVD risk prediction, then the prediction can be made using available parameters, but the accuracy is modest resulting in too many false positives and false negatives
Solution Approach 1:
The patent segments the CVD risk prediction process into multiple specialized deep learning models, each trained to detect specific retinal features associated with different CVD risk factors (blood pressure, cholesterol, glycaemic control). This segmentation allows each model to focus on specific patterns, improving overall prediction accuracy and reducing false positives compared to single-label training approaches.
Solution Approach 2:
The patent transitions from training against single labels (chronological age or conventional equations) to multi-dimensional training using multiple independent labels representing different CVD risk factors. This dimensional expansion allows the model to capture complex relationships between retinal features and various risk contributors simultaneously, improving prediction precision and reliability.
2Device complexity
If retinal images are trained against single labels (chronological age or conventional equations), then the training process is simplified, but the ability to identify major contributors to CVD risk is lost
Solution Approach 1:
The training process is segmented into multiple independent training tasks, with each model trained against a specific label corresponding to a CVD risk factor (e.g., blood pressure, cholesterol). This segmentation preserves information about major contributors while maintaining manageable training complexity through modular model development and independent loss functions for each risk factor.
Solution Approach 2:
The deep learning system is designed with multi-functionality, where the same retinal image input can be processed by multiple models to extract information about different CVD risk factors. This universal approach allows a single imaging system to provide comprehensive risk assessment across multiple dimensions without requiring separate imaging procedures.
3Ease of operation
If conventional CVD risk equations are used, then treatment decisions can be made based on predicted risk, but the modest accuracy prevents better targeting of medications and treatment programs
Solution Approach 1:
The patent applies local quality by providing differentiated risk assessments for different CVD risk factors (blood pressure, cholesterol, glycaemic control) rather than a single aggregate score. This allows clinicians to identify which specific risk factors contribute most to an individual's overall risk, enabling precise targeting of medications and treatment programs to address specific modifiable risk factors.
Solution Approach 2:
The system transforms the conventional single-parameter CVD risk score into multiple parameters representing different risk factors. This parameter expansion allows for more nuanced risk stratification and treatment decision-making, where clinicians can adjust treatment based on which specific parameters (risk factors) are elevated, improving the precision of treatment targeting.
Data Source
AI summary
Systems and methods for determining one or more recommendations for management of wellbeing of an individual are disclosed. An indication of relative cardiovascular aging of the individual is determined based at least in part on a predicted risk of cardiovascular disease (CVD) of the individual determined by a deep learning model based on one or more fundus images. The recommendations for management of the individual's wellbeing are based at least in part on a determined difference between the actual chronological age of the individual and an indication of relative cardiovascular aging of the individual.