Blood cell image segmentation and classification: a systematic review.

Area of Science:

• Medical Image Analysis
• Computational Pathology
• Hematology

Background:

• Blood diseases (leukemia, anemia, lymphoma, thalassemia) involve abnormalities in white blood cells (WBC) and red blood cells (RBC).
• Accurate diagnosis relies on hematologist expertise, with growing interest in computer-aided diagnostic (CAD) techniques.
• Existing surveys lack a holistic view, focusing narrowly on specific aspects like segmentation or classification.

Purpose of the Study:

• To provide a comprehensive and systematic review of blood image analysis using deep learning.
• To focus on medical image processing and deep learning for WBC and RBC morphological characterization.
• To cover segmentation, classification, feature selection, evaluation parameters, and dataset selection.

Main Methods:

• Systematic literature review of deep learning techniques in blood image analysis.
• Analysis of segmentation, classification, feature selection, evaluation matrices, and dataset selection methodologies.
• Focus on morphological characterization of WBCs and RBCs.

Main Results:

• Researchers often manually acquire images (50% for WBC segmentation, 60% for RBC segmentation).
• The ALL-IDB dataset is common for WBC classification (45%), while RBC classification frequently uses manually acquired images (73%).
• Morphological features are preferred for classification (55% for WBC, 80% for RBC).

Conclusions:

• CAD techniques, particularly deep learning, can enhance blood disease diagnosis.
• Inconsistent dataset selection highlights the need for standardized, high-quality datasets.
• Future research should explore and innovate with morphological features and standardized datasets.