Machine-Learning Risk Factor Selection for PCI Target Lesion Failure
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current risk prevention solutions for percutaneous coronary intervention (PCI) target lesion failure (TLF) are based on outdated clinical parameters and fail to accurately predict future TLF events due to imbalanced patient data.
Innovation Solution
A computing system that identifies minority and majority class data in patient-level data, oversamples the minority class data to create synthetic class data, and reduces a set of risk factor variables using a machine learning classifier, such as a multi-layer neural network, Random Forest, or Extra Trees classifier, to balance the data and identify the most relevant risk factors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If simple clinical parameters from outdated studies are used for risk prediction, then the solution is easy to implement, but the accuracy of predicting TLF events deteriorates
Solution Approach 1:
The patent transforms the input data from simple clinical parameters to comprehensive patient-level data including laboratory results, imaging data, and procedural details. It also changes the data structure from aggregated to patient-level granularity, enabling machine learning models to process and identify patterns that simple parameters cannot capture, thereby improving prediction accuracy while maintaining implementation feasibility through automated processing.
2Measurement precision
If extensive patient-level data is collected to improve prediction accuracy, then the prediction accuracy improves, but the data collection complexity and time increase
Solution Approach 1:
The patent performs preliminary data processing by pre-processing patient-level data from multiple sources before analysis. The machine learning model is trained in advance to identify the most relevant risk factor variables, so when deployed, it can quickly process new patient data without requiring extensive real-time data collection. The system pre-identifies key risk factors from comprehensive data, reducing the actual data collection time needed at the point of care.
3Adaptability or versatility
If a large set of risk factor variables is analyzed to capture all potential risks, then the comprehensiveness of risk assessment improves, but the complexity of the analysis system increases
Solution Approach 1:
The patent extracts and identifies the most relevant risk factor variables from a comprehensive set of patient-level data using machine learning classification. Instead of analyzing all available variables, the system automatically selects and extracts only the most significant risk factors for each patient, reducing analysis complexity while maintaining comprehensive risk assessment. This extraction approach allows the system to handle diverse data types without requiring complex manual analysis of every variable.
4Reliability
If minority class data (patients with health failure) is used for training, then the model learns from actual failure cases, but the imbalanced data distribution reduces model performance
Solution Approach 1:
The patent creates synthetic copies of minority class data through data augmentation techniques. By generating synthetic patient records that resemble actual health failure cases, the system increases the representation of minority class samples in the training dataset. This copying approach allows the machine learning model to learn from sufficient examples of health failure cases without requiring an imbalanced amount of real-world failure data, thereby improving model performance while maintaining reliability.
Data Source
AI summary
A system includes: a processing circuit including a memory device coupled to a processor, the memory device configured to store instructions thereon that, when executed by the processor, cause the processor to: generate synthetic class data using minority class data to obtain balanced class data including the minority class data corresponding to patients with a health failure, the synthetic class data corresponding to the health failure, and majority class data corresponding to patients without the health failure; automatically reduce, using a machine learning classifier, risk factor variables for the health failure to a reduced set of risk factor variables based on the balanced class data; and execute the machine learning classifier using as input a reduced set of risk factor variable data for a patient corresponding to the reduced set of risk factor variables to generate a probability indicator of the health failure for the patient.