Abstract: The use of machine learning (ML) methods has become very popular in transfusion medicine because it can provide new ways of diagnosis, prognosis, and prediction of responses to treatment when dealing with blood cancer. In this review, we examine the literature on ML-based technologies in HMs. Supervised learning techniques like Support Vector Machines (SVM) and Random Forests (RF) have been used to subtype leukemia and lymphomas based on high-dimensional gene expression data, where each sample xi=(xi1, xi2,….xid) is a d-dimensional vector of gene expression levels (one number per gene). A function f:Rd →{0,1} is a model that learned class labels (0 = healthy, 1 = diseased), any such vector, predicted class label A, big accuracy for cancer class distinctions, i.e., cancer subtypes equal accuracy. These models have turned up novel biomarkers xjx significantly linked to patient survival times at the time that enable physicians to better stratify risk and individualize their therapy plan approaches. Yet, challenges are still on the way despite this promising attack. In addition, there are major challenges related to data heterogeneity, low internal comparability due to a lack of a standardised pre-processing pipeline, low external validation, and ethical challenges (e.g., bias, model transparency), which prevent full clinical integration. Concerning methods for ensuring privacy preservation and mitigating bias in collaborative learning, rigorous cross-validation techniques, federated learning models, and trusted ethical codes for AI development will play a prominent role in the future.

Keywords: blood cancer, machine learning, diagnosis, prognosis prediction, treatment response prediction, biomarker

Pages: 90-101