IoT Application of Transfer Learning in Hybrid Artificial Intelligence Systems for Acute Lymphoblastic Leukemia...

Area of Science:

• Medical Diagnostics
• Artificial Intelligence in Healthcare
• Computational Biology

Background:

• Accurate white blood cell classification is crucial for diagnosing pediatric acute lymphoblastic leukemia (ALL).
• Current diagnostic methods rely heavily on manual microscopic examination of bone marrow, necessitating efficient automated classification tools.
• The need for computationally efficient AI models suitable for Internet of Things (IoT) devices in healthcare settings is growing.

Purpose of the Study:

• To develop and evaluate an optimized, IoT-friendly hybrid artificial intelligence (AI) system for accurate white blood cell classification.
• To leverage transfer learning with pre-trained neural networks combined with machine learning algorithms for enhanced diagnostic performance.
• To demonstrate the efficacy of hybrid AI in resource-constrained environments for medical diagnosis.

Main Methods:

• Implementation of a hybrid AI system utilizing a MobileNet v2 encoder pre-trained on the ImageNet dataset.
• Integration of machine learning algorithms (XGBoost, Random Forest, Decision Tree) as the classification head.
• Training and validation of the system on medical datasets for white blood cell classification.

Main Results:

• The developed hybrid AI system achieved an average classification accuracy exceeding 90%, with a peak accuracy of 97.4%.
• The system demonstrated high classification performance, validating the effectiveness of hybrid AI for tasks with low computational complexity.
• Transfer learning proved effective in enhancing the performance of the AI model for medical image analysis.

Conclusions:

• Hybrid AI systems, combining deep learning encoders with traditional machine learning classifiers, offer a powerful and accurate tool for blood cell classification.
• The proposed IoT-friendly architecture is suitable for deployment in various medical settings, potentially improving diagnostic workflows.
• The demonstrated accuracy suggests this methodology can be extended to aid in the diagnosis of other hematological diseases, supporting clinical decision-making and treatment planning.