Self-Supervised Learning-Based General Laboratory Progress Pretrained Model for Cardiovascular Event Detection.

Area of Science:

• Machine learning applications in healthcare
• Biomedical data analysis
• Translational bioinformatics

Background:

• Patient data in disease care presents challenges like data irregularity and sparsity, especially in longitudinal studies.
• Rare disease cases often have limited patient data and episodic observations, hindering predictive modeling.
• Machine learning (ML) offers potential for leveraging patient data but requires robust models to handle data complexities.

Purpose of the Study:

• To develop a generalized laboratory progress (GLP) model using self-supervised learning (SSL) for cardiovascular (CV) laboratory markers.
• To transfer knowledge from the GLP model trained on prevalent CV cases to aid in detecting specific CV events.
• To overcome data limitations in rare or specific disease cases through knowledge transfer.

Main Methods:

• Employed a two-stage training approach for the GLP model, incorporating interpolated data to enhance SSL performance.
• Utilized self-supervised learning (SSL) to pretrain the GLP model on six common laboratory markers in prevalent cardiovascular cases.
• Transferred the pretrained GLP model for the specific task of target vessel revascularization (TVR) detection.

Main Results:

• The two-stage training approach significantly improved upon pure SSL performance.
• The generalized laboratory progress (GLP) model demonstrated strong transferability for downstream tasks.
• Classification accuracy for cardiovascular event detection improved from 0.63 to 0.90 after GLP processing, with substantial superiority in all metrics.

Conclusions:

• The study successfully demonstrated translational engineering by transferring patient progression data between cohorts.
• The transferability of disease progression models optimizes examination and treatment strategies, improving patient prognosis.
• This approach shows promise for application to other diseases, enhancing diagnostic and prognostic capabilities using common laboratory parameters.