An improved multiclass classification of acute lymphocytic leukemia using enhanced glowworm swarm optimization

Area of Science:

• Medical Imaging
• Computational Biology
• Hematology

Background:

• Acute Lymphoblastic Leukemia (ALL) is a prevalent pediatric blood cancer characterized by immature white blood cell proliferation.
• Current diagnostic methods for ALL, such as bone marrow aspiration, are often time-consuming, error-prone, and require expert interpretation.
• There is a need for efficient and accurate automated systems to aid in ALL classification.

Purpose of the Study:

• To develop a computer-aided automatic classification system for differentiating Benign, Early, Pro-B, and Pre-B ALL subtypes.
• To enhance diagnostic accuracy and efficiency in ALL detection.

Main Methods:

• Image pre-processing and Region of Interest extraction using a Multilevel Hierarchical Marker-Based Watershed Algorithm (MHMW).
• Feature selection employing the Enhanced Glowworm Swarm Optimization (EGSO) algorithm.
• Multiclass classification using Decision Tree, Random Forest, Multi-Layer Perceptron, Naive Bayes, and Support Vector Machine kernels.

Main Results:

• The integrated MHMW and EGSO approach, coupled with Random Forest, demonstrated superior performance.
• The system achieved high accuracy (98.23%), precision (98.25%), and F1 score (98.23%) in classifying ALL subtypes.
• Performance was benchmarked against other optimization algorithms like Particle Swarm Optimization, Artificial Bee Colony Optimization, and Elephant Herd Optimization.

Conclusions:

• The proposed computer-aided system effectively classifies ALL subtypes with high accuracy.
• This automated approach offers a potential improvement over traditional diagnostic methods for ALL.
• The integration of MHMW and EGSO algorithms shows significant promise for medical image analysis in hematological malignancies.