Method for Diagnosis of Acute Lymphoblastic Leukemia Based on ViT-CNN Ensemble Model

Area of Science:

• Computational Biology and Bioinformatics
• Medical Imaging and Diagnostics
• Artificial Intelligence in Healthcare

Background:

• Acute lymphocytic leukemia (ALL) is a significant cause of cancer in adults and children.
• Distinguishing leukemic B-lymphoblast cells from normal B-lymphoid precursors via microscopy is challenging due to morphological similarities.
• Accurate and timely diagnosis is crucial for effective ALL treatment and improved survival rates.

Purpose of the Study:

• To develop an advanced computational model for accurate classification of cancer cells in acute lymphoblastic leukemia (ALL) images.
• To assist clinicians in the diagnosis of ALL by providing a reliable computer-aided method.
• To address challenges posed by unbalanced datasets and noise in cell image analysis.

Main Methods:

• Proposed a ViT-CNN ensemble model, integrating Vision Transformer (ViT) and Convolutional Neural Network (CNN) for feature extraction.
• Developed a Difference Enhancement-Random Sampling (DERS) data augmentation technique for unbalanced and noisy datasets.
• Employed a symmetric cross-entropy loss function to mitigate the impact of data noise.

Main Results:

• The ViT-CNN ensemble model achieved a classification accuracy of 99.03% on the test dataset.
• Experimental comparisons demonstrated superior performance of the proposed model over other existing methods.
• The DERS method successfully created a balanced dataset, enhancing model robustness.

Conclusions:

• The ViT-CNN ensemble model effectively differentiates between leukemic and normal cells, offering a promising tool for ALL diagnosis.
• The proposed data augmentation and loss function strategies improve classification performance on challenging datasets.
• This AI-driven approach represents a significant advancement in computer-aided diagnosis for acute lymphoblastic leukemia.