Blood Cancer Detection Based on CNN and Integrated Learning

Acute lymphoblastic leukemia (ALL), one of the most common hematologic malignancies, emphasizes the need for early detection and accurate diagnosis to improve patient prognosis. Traditional diagnostic methods of manual microscopy are subjective, time-consuming, and require specialized expertise. Current artificial intelligence (AI) methods for acute lymphoblastic leukemia detection face challenges in terms of subtype classification accuracy and data generalization ability. To address these limitations, this study proposes a CNN architecture, which is a model trained using stochastic gradient descent (SGD) and to extract hierarchical features and reduce overfitting while improving efficiency. The dataset for this study contains 3,242 PBS images from 89 patients with suspected ALL, which were preprocessed using normalization, and classification coding. Evaluated on the dataset, the proposed model achieves an impressive accuracy of $96 \%$, showing significant performance in all subtypes (Malignant_early Pre-B, Pre-B, Pro-B) and benign samples, especially for Malignant_Pro-B ($100 \%$ recall) and Malignant_Pre- B ($99 \%$ accuracy). This method provides a robust and efficient solution for automated diagnosis of acute lymphoblastic leukemia (ALL) in peripheral blood smear (PBS) images for clinical applications.