Indirect reference interval estimation using a convolutional neural network with application to cancer antigen 125.

Area of Science:

• Biomedical data analysis
• Machine learning in diagnostics
• Clinical laboratory science

Background:

• Indirect methods for reference interval (RI) estimation accelerate RI establishment for improved clinical assessments.
• Pathological patient data in routine testing datasets require sophisticated analysis methods.
• Accurate RIs are crucial for effective disease diagnosis and monitoring.

Purpose of the Study:

• To develop a novel convolutional neural network (CNN) model for estimating reference intervals (RIs) from routine pathology data.
• To generate synthetic data for training the CNN model to identify healthy distributions within pathological admixtures.
• To evaluate the CNN model's performance against state-of-the-art methods and demonstrate its real-world applicability.

Main Methods:

• Development of a novel convolutional neural network (CNN) model.
• Generation of synthetic data for training and validation.
• Evaluation using the RlBench benchmark and a real-world dataset (CA-125).

Main Results:

• The developed CNN model significantly outperformed the state-of-the-art method (refineR) on the RlBench benchmark.
• The model successfully identified underlying healthy distributions within pathological data admixtures.
• Age-specific reference intervals for cancer antigen 125 (CA-125) were estimated, revealing strong age-dependency.

Conclusions:

• The novel CNN model provides a robust and accurate method for indirect reference interval estimation.
• The model's ability to handle pathological data improves the reliability of RIs derived from routine testing.
• Estimated age-specific CA-125 RIs have significant implications for ovarian cancer diagnostics.