Leukemia classification using the deep learning method of CNN.

Area of Science:

• Medical Imaging Analysis
• Computational Pathology
• Artificial Intelligence in Healthcare

Background:

• Manual examination of low-intensity medical images (LI-MI) is challenging due to varied outcomes and time constraints.
• Accurate classification of leukemia from medical images is crucial for timely diagnosis and treatment.

Purpose of the Study:

• To enhance the quality of low-intensity medical images.
• To accurately classify leukemia using a deep convolutional neural network (DCNN) strategy.

Main Methods:

• Employed DCNN models (ResNet-34 & DenseNet-121) for leukemia classification.
• Applied histogram equalization-based adaptive gamma correction and guided filtering to improve image intensity and preserve details.
• Utilized ASH and ALL-IDB datasets comprising 1,079 images (779 leukemia positive, 300 normal) for validation.

Main Results:

• Achieved high classification accuracies of 99.2% and 98.4% on the datasets.
• Reported excellent specificity, sensitivity, and precision rates, exceeding 98% for both datasets.
• Demonstrated superior performance compared to other machine learning classifiers like SVM, Decision Tree, Naive Bayes, Random Forest, and VGG-16.

Conclusions:

• The proposed DCNN strategy effectively improves low-intensity medical image quality.
• The DCNN approach offers superior accuracy for leukemia classification compared to existing machine learning methods.
• This study highlights the potential of DCNNs in advancing diagnostic capabilities for leukemia.