Machine Learning for the Interpretation of Serum Protein and Immunofixation Electrophoresis in Multiple Myeloma: A...

Background: Serum protein electrophoresis (SPE) and immunofixation electrophoresis (IFE) are essential in diagnosing multiple myeloma (MM). Machine learning (ML) methods have emerged as a potential approach to the automation process of electrophoresis interpretation to improve diagnostic efficiency and improve turnaround time. This scoping review mapped current applications of ML in the diagnosis and classification of MM, with emphasis on automated SPE and IFE interpretation. Methods: PubMed, Web of Science, and Scopus databases were searched following PRISMA-ScR guidance. Articles published between January 2015 and April 2025 applying ML to MM diagnosis were included, and findings were synthesised descriptively and thematically. Results: Thirteen studies were included, identifying two dominant application domains: ML-based analysis of IFE images and SPE patterns. Deep learning models, particularly convolutional neural networks, achieved expert-level performance in IFE analysis, with peak accuracies reaching 99.82% and F1-scores frequently exceeding 0.90. For SPE interpretation, both deep learning and classical tree-based models demonstrated robust diagnostic yields, achieving Areas Under the Curve (AUC) between 0.90 and 0.99 and overall accuracies ranging from 82.8% to 99.1%. Notably, top-performing models recorded F1-scores up to 0.98, and in some instances, they surpassed the precision of human expert panels. Conclusions: Explainable AI methods were increasingly incorporated, though external validation, multimodal integration, explainability, and real-world deployment were limited. ML demonstrates potential to augment SPE and IFE interpretation in MM diagnostics. Future work should focus on multicentre validation, explainability, and clinical integration to support safe implementation.