Serum Protein Electrophoresis (SPE) is a biochemical routine analysis for the relative quantification of serum proteins. While the primary indication for the analysis nowadays is the detection of monoclonal gammopathy, other pathological conditions such as an acute or chronic inflammatory process or nephrotic syndrome in the form of typical dys-proteinemias are also visible.
The current standard method for SPE is capillary zone electrophoresis. In this process, serum proteins are separated in a liquid, alkaline medium in a capillary using high-volt-age. Subsequent quantification is done through UV detection. The resulting measure-ment curve is then divided into individual protein fractions (Albumin, α1-, α2-, β1-, β2-, and γ-Globulin) and, if necessary, additional bands indicative of gammopathy. Current programs are using mathematical methods, such as the identification of minima/maxima, for the separation of fractions. The curves are then usually manually validated, and the fractionation may be adjusted if needed.
While the manual interpretation of a significant gammopathy is straightforward, the inter-pretation of weak gammopathies and other dysproteinemias requires a high level of ex-perience and is still characterized by some uncertainty and subjectivity. Chabrun et al. demonstrated a possible new interpretation approach using deep learning for a reliable classification and result interpretation.
Building on this, we aim to further refine the machine-learning-based classification and report commentary of serum electrophoresis, as well as validate its application in routine diagnostics. Additionally, we prospectively aim to biochemically validate SPE-morpho-logical dysproteinemias based on quantification of surrogate parameters. With success-ful establishment and validation of the methodology, there is the prospect of increased automation of the existing process and improved report quality in the interpretation of borderline result constellations.
Chabrun, F., Dieu, X., Ferre, M., Gaillard, O., Mery, A., Chao de la Barca, J. M., Taisne, A., Urbanski, G., Reynier, P., & Mirebeau-Prunier, D. (2021). Achieving Expert-Level In-terpretation of Serum Protein Electrophoresis through Deep Learning Driven by Human Reasoning. Clinical chemistry, 67(10), 1406–1414. https://doi.org/10.1093/clinchem/hvab133
