Incremental learning for acute lymphoblastic leukemia classification based on hybrid deep learning using blood smear...

Area of Science:

• Medical Imaging
• Computational Biology
• Hematology

Background:

• Leukemia is a common blood cancer characterized by irregular immature malignant cell production in bone marrow.
• Early detection and treatment are crucial for managing leukemia, as it compromises the immune system and can be fatal.
• Manual microscopic analysis of blood smears for leukemia detection is time-consuming and labor-intensive.

Purpose of the Study:

• To develop an automated and precise method for classifying acute lymphoblastic leukemia (ALL) using microscopic blood smear images.
• To introduce an incremental learning model, TSCO-L-LeNet, for enhanced leukemia diagnosis.
• To improve the speed, accuracy, and safety of leukemia diagnosis.

Main Methods:

• Image preprocessing using an adaptive median filter and segmentation with Scribble2label.
• Image augmentation and feature extraction from segmented images.
• Classification of ALL using Long Short-Term Memory-LeNet (L-LeNet) with incremental learning, optimized by Tangent Sand Cat Swarm Optimization (TSCO) for weight training.

Main Results:

• The TSCO-L-LeNet model achieved high performance metrics.
• Achieved an accuracy of 0.987, True Negative Rate (TNR) of 0.977, and precision of 0.979.
• Demonstrated a low False Negative rate of 0.033 and False Positive rate of 0.023, with a recall of 0.967.

Conclusions:

• The proposed TSCO-L-LeNet model provides a highly accurate and efficient automated system for acute lymphoblastic leukemia classification.
• Utilizing blood smear images with this model significantly reduces diagnosis time and enhances accuracy.
• The approach offers a cheaper, faster, and safer alternative for leukemia diagnosis services.