An efficient decision support system for leukemia identification utilizing nature-inspired deep feature optimization.

Area of Science:

• Medical Imaging and Diagnostics
• Computational Pathology
• Artificial Intelligence in Healthcare

Background:

• Leukemia, particularly acute lymphoblastic leukemia (ALL), requires rapid and accurate diagnosis for effective treatment due to its aggressive nature.
• Computer vision for leukemia diagnosis faces challenges from the complex morphology of blood cells and the high computational demands of deep neural networks.
• Existing methods often struggle with feature extraction and selection, impacting diagnostic accuracy and efficiency.

Purpose of the Study:

• To develop an efficient computational pipeline for the binary and subtype classification of acute lymphoblastic leukemia (ALL) using medical images.
• To enhance the clarity and discriminability of blood cell images through a novel neighborhood pixel transformation method.
• To improve feature extraction and selection efficiency for deep learning models in ALL diagnosis.

Main Methods:

• A novel neighborhood pixel transformation technique utilizing differential evolution was employed to preprocess blood cell images.
• A hybrid feature extraction approach combined transfer learning from InceptionV3 and DenseNet201 deep neural network models.
• A customized binary Grey Wolf Algorithm was used for feature selection, achieving an 80% reduction in feature size while retaining critical information.

Main Results:

• The pipeline achieved a 98.1% accuracy, 98.1% sensitivity, and 98% precision for binary classification of ALL on public datasets.
• For ALL subtype classification, the best performance reached 98.14% accuracy, 78.5% sensitivity, and 98% precision.
• The proposed feature selection method demonstrated superior convergence compared to traditional meta-heuristics and achieved comparable or better performance than existing techniques.

Conclusions:

• The developed efficient pipeline offers a promising approach for accurate and rapid diagnosis of acute lymphoblastic leukemia (ALL).
• The combination of image transformation, hybrid feature extraction, and optimized feature selection significantly enhances classification performance.
• This AI-driven method holds potential for improving clinical decision support systems in hematological diagnostics.