Development and Validation of a Protein Electrophoresis Classification Algorithm: Tabular Data-Based Alternative.

Area of Science:

• Clinical Chemistry
• Medical Laboratory Science
• Machine Learning in Healthcare

Background:

• Serum protein electrophoresis (SPE) is a standard diagnostic tool.
• Current SPE interpretation relies on subjective visual assessment of electropherograms by medical laboratory scientists.
• This visual method can be time-consuming and prone to inter-observer variability.

Purpose of the Study:

• To develop and evaluate an efficient machine learning approach for SPE interpretation.
• To provide a robust and interpretable alternative to traditional image-based deep learning methods.
• To leverage tabular numerical data from SPE profiles for automated analysis.

Main Methods:

• Developed a machine learning model utilizing tabular numerical data derived from SPE profiles.
• Focused on a data-driven approach rather than image-based deep learning.
• Ensured the model's interpretability for clinical application.

Main Results:

• The proposed method efficiently processes numerical SPE data.
• It offers a robust alternative to current visual interpretation techniques.
• The approach provides enhanced interpretability compared to complex deep learning models.

Conclusions:

• Tabular data-based machine learning presents a viable and efficient method for SPE interpretation.
• This approach offers a practical and interpretable alternative for medical laboratory scientists.
• Future work can focus on further validating and integrating this method into routine clinical practice.