Machine Learning Applications in Blood Quality Assessment: A Systematic Literature Review

Abstract

This review article aims to evaluate the machine learning algorithms used for blood packet quality prediction, compare imaging and biophysical techniques used for non-invasive evaluation of red blood cell quality, understand the impact of donor-related and storage-related factors on blood cell quality, compare label-free and labelled models, and understand the possibility of introducing such models in clinical practice. The systematic study of research conducted in three continents (North America, Europe and Asia) revealed that machine learning models were well predictive (76.7-98%) of RBC morphology and biochemical markers and generally more accurate than expert judgments when applied to imaging flow cytometry, quantitative phase imaging and microfluidics. Labels and automated classification allow objective and reproducible assessment and measurement of donor and storage variability in RBC quality in high-throughput imaging. However, there is a lack of datasets and prospective validation limits generalisation and clinical implementation. The complexity of equipment, lack of standardization, and the need to adapt to workflows are among the challenges of integration. Overall, these findings point to potential advances in transfusion safety and the effectiveness of non-invasive machine learning systems, with the need to broaden and standardise the validation of such systems. The technologies that will be developed in the future will benefit the operation of blood banks and allow for a more personalized approach to blood transfusion organization.

The authors searched Scopus and Web of Science (SCI) using predefined keywords, screened 1,234 records, and included 58 studies in the review. The review followed a structured screening and quality-assessment methodology, with duplicate removal, independent reviewer screening, and narrative synthesis of the included evidence.

A major gap remains in the lack of standardized and validated methods for comparing blood cell quality assessment approaches