Computer-Aided Diagnosis System for Blood Diseases Using EfficientNet-B3 Based on a Dynamic Learning Algorithm.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Hematology

Background:

• Leukemia, particularly acute lymphoblastic leukemia (ALL), is a common blood cancer and a leading cause of death, necessitating early and accurate detection for improved outcomes.
• Computer-aided diagnostic (CAD) models offer a promising approach for the early and effective detection of leukemia.
• The C-NMC_Leukemia dataset is utilized for evaluating diagnostic models.

Purpose of the Study:

• To propose and evaluate an automated classification model for the early detection of acute lymphoblastic leukemia (ALL).
• To develop a model capable of automatically adjusting its learning rate for enhanced performance.
• To assess the model's efficacy in identifying malaria parasites in microscopic blood images.

Main Methods:

• A classification model based on the EfficientNet-B3 convolutional neural network (CNN) was developed.
• The model incorporated an automated learning rate adjustment mechanism, comparing loss value and training accuracy each epoch.
• The model was trained and evaluated on the pre-processed C-NMC_Leukemia dataset.

Main Results:

• The proposed EfficientNet-B3 model achieved high performance in ALL detection, with average precision of 98.29%, recall of 97.83%, specificity of 97.82%, accuracy of 98.31%, and DSC of 98.05%.
• The model also demonstrated strong performance in malaria parasite identification, with average precision, recall, specificity, accuracy, and DSC all at 97.68%.
• The proposed model outperformed existing classifiers in the evaluated metrics.

Conclusions:

• The developed automated EfficientNet-B3 model shows significant potential for accurate and early detection of acute lymphoblastic leukemia (ALL).
• The model's adaptability, including its custom learning rate adjustment, contributes to its superior performance.
• The model's versatility is further highlighted by its successful application in malaria parasite detection, indicating broad applicability in medical image analysis.