A machine-learned predictor of colonic polyps based on urinary metabolomics

Area of Science:

• Biochemistry
• Medical Diagnostics
• Machine Learning

Background:

• Colorectal cancer screening is crucial for early detection.
• Current screening methods like fecal tests have limitations.
• Non-invasive diagnostic tools are needed to improve patient compliance and accuracy.

Purpose of the Study:

• To develop and validate an automated diagnostic test for identifying patients who require a colonoscopy.
• To assess the test's ability to distinguish between patients needing and not needing colonoscopy based on urine NMR spectra.
• To evaluate the flexibility of the test in adjusting sensitivity and specificity.

Main Methods:

• Utilized proton nuclear magnetic resonance ((1)H-NMR) spectroscopy on single spot urine samples.
• Quantified metabolic profiles using targeted profiling.
• Employed machine learning algorithms to analyze metabolic profiles, colonoscopy outcomes, and medical histories.
• Developed a classifier to predict the need for colonoscopy in average or above-average risk individuals.

Main Results:

• The developed classifier achieved a sensitivity of 64% and a specificity of 65%.
• The system demonstrated the ability to adjust the tradeoff between sensitivity and specificity.
• Successfully distinguished between patients who required colonoscopy and those who did not, based on urine metabolic profiles.

Conclusions:

• An automated urine NMR-based test can accurately aid in identifying patients requiring colonoscopy.
• This non-invasive approach offers a flexible alternative to current colorectal cancer screening methods.
• Further refinement of the machine learning classifier may enhance diagnostic performance.