Deep Learning in Hematology: From Molecules to Patients.

Area of Science:

• Computational Biology
• Medical Informatics
• Hematology

Background:

• Deep learning (DL), a subset of machine learning, demonstrates significant advancements in medicine.
• Hematology applications of DL range from fundamental molecular research to clinical patient management.
• Understanding DL basics, architectures, and training is crucial for its medical implementation.

Purpose of the Study:

• To review the diverse applications of deep learning in hematology.
• To analyze DL models' architecture, performance, and limitations in hematological contexts.
• To provide an accessible introduction to DL for non-experts in the field.

Main Methods:

• Review of existing literature on deep learning applications in hematology.
• Analysis of DL model architectures, performance metrics, and limitations.
• Categorization of applications across molecular, cellular, tissue, and patient levels.

Main Results:

• DL enhances molecular analysis (multi-omics, protein structure) and cellular diagnostics (cytomorphology, flow cytometry, whole slide imaging).
• Large language models (LLMs) enable analysis of clinical data, electronic health records, and clinical notes.
• Promising results are observed, but challenges in model generalizability and explainability remain.

Conclusions:

• Deep learning offers transformative potential across hematology, improving diagnostics and data analysis.
• Integration of DL in hematology lags behind other medical fields, necessitating further research and development.
• Addressing challenges in generalizability and explainability is key for broader clinical adoption.