Optimizing a Deep Residual Neural Network with Genetic Algorithm for Acute Lymphoblastic Leukemia Classification.

Area of Science:

• Medical Imaging
• Computational Biology
• Artificial Intelligence in Healthcare

Background:

• Acute lymphoblastic leukemia (ALL) is a prevalent childhood cancer originating in bone marrow.
• Microscopy image analysis aids ALL diagnosis but is subjective and time-consuming.
• Automated analysis via computer vision offers potential for efficient and objective screening.

Purpose of the Study:

• To develop and evaluate a hybrid model for accurate ALL prediction using microscopy images.
• To optimize model hyperparameters using a genetic algorithm (GA).
• To compare GA hyperparameter optimization with Random Search and Bayesian optimization.

Main Methods:

• A hybrid model integrating a genetic algorithm (GA) with the ResNet-50V2 residual convolutional neural network (CNN) was proposed.
• The model was trained and validated on the ALL-IDB dataset.
• Hyperparameter optimization was performed using GA, Random Search, and Bayesian optimization.

Main Results:

• The GA-optimized hybrid model achieved a high accuracy of 98.46% for ALL prediction.
• GA-based hyperparameter tuning outperformed Random Search and Bayesian optimization in enhancing classifier accuracy.
• The study demonstrates the efficacy of GA in optimizing deep learning models for medical image analysis.

Conclusions:

• The proposed GA-optimized ResNet-50V2 model offers a highly accurate and efficient method for diagnosing ALL from microscopy images.
• This computer vision approach presents a promising alternative for real-world leukemia screening applications.
• Genetic algorithm optimization is effective for improving the performance of deep learning models in medical diagnostics.