Quantification of human sperm concentration using machine learning-based spectrophotometry

Area of Science:

• Biomedical Engineering
• Spectroscopy
• Artificial Intelligence

Background:

• Spectrophotometry is a non-invasive method for material analysis based on light absorption.
• Accurate quantification of human sperm concentration is crucial for male infertility diagnosis.
• Current methods for semen analysis have limitations in speed and cost.

Purpose of the Study:

• To apply artificial neural networks (ANN) to spectrophotometry for quantifying human sperm concentration.
• To develop and validate a full spectrum neural network (FSNN) model for this application.
• To establish a rapid, low-cost, and accurate method for sperm concentration measurement.

Main Methods:

• Development of a well-trained full spectrum neural network (FSNN) model.
• Examination of sperm sample absorption responses across a light spectra range (390-1100 nm).
• Testing the FSNN model on samples from 41 human subjects.

Main Results:

• The FSNN model accurately estimates sperm concentration with over 93% prediction accuracy.
• The method achieved 100% agreement with clinical assessments in differentiating healthy donor from patient samples.
• The developed technique demonstrates superior performance compared to existing spectrophotometry methods for semen analysis.

Conclusions:

• Machine learning-based spectrophotometry with FSNN is a powerful technique for sperm quantification.
• This approach offers a rapid, low-cost alternative for semen analysis.
• The study presents novel research and clinical opportunities for addressing male infertility.