An Aggregated-Based Deep Learning Method for Leukemic B-lymphoblast Classification.

Area of Science:

• Hematology
• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Leukemia is a bone marrow cancer affecting children and adolescents, characterized by abnormal lymphocyte growth.
• Early diagnosis is critical for curability, but increasing global incidence necessitates improved detection methods.
• Computer-aided diagnosis (CAD) models offer potential for assisting clinicians in early leukemia detection.

Purpose of the Study:

• To develop an aggregated deep learning model for accurate classification of Leukemic B-lymphoblasts.
• To enhance the reliability and accuracy of automated leukemia detection systems.
• To address the challenge of limited dataset size in medical AI development.

Main Methods:

• Employed data augmentation techniques to overcome dataset limitations.
• Utilized a transfer learning strategy to accelerate model training and improve performance.
• Developed an aggregated deep learning model by fusing features from multiple networks.

Main Results:

• The proposed aggregated deep learning model achieved a test accuracy of 96.58% for Leukemic B-lymphoblast diagnosis.
• The model successfully fused features from various deep learning networks, outperforming individual models.
• Data augmentation and transfer learning effectively improved the deep learner's performance.

Conclusions:

• The developed aggregated deep learning model demonstrates high accuracy in diagnosing Leukemic B-lymphoblasts.
• This AI-driven approach shows significant promise for computer-aided diagnosis in early leukemia detection.
• The study highlights the effectiveness of combining data augmentation and transfer learning for robust medical image analysis.