Uso de redes neurais convolucionais no diagnóstico de leucemia linfoblástica aguda

Abstract

Acute Lymphoblastic Leukemia (ALL) is an aggressive hematologic cancer that predominantly affects children and requires early diagnosis to improve survival rates. Traditional diagnostic methods, such as bone marrow aspiration and flow cytometry, are invasive, expensive, and often inaccessible in resource-limited settings (Pui et al., 2015). Studies such as (Ghaderzadeh et al., 2022) propose the use of Artificial Intelligence (AI) models to assist physicians in identifying the disease through imaging, focusing on the differentiation between benign cells (hematogones) and malignant lymphoblasts. However, these approaches are often complex and resource-intensive, including longer processing times. This monograph also proposes the use of Convolutional Neural Networks (CNNs) to support the screening and diagnosis of ALL from Peripheral Blood Smear (PBS) images, employing the EfficientNet-B3 architecture, known for balancing performance and efficiency (Tan e Le, 2019). Supported by the FastAI library, a classification algorithm was implemented and achieved an accuracy of 98.92% on the test set. The results were compared with those of the reference study, which used DenseNet201 combined with HSV color segmentation, indicating that the proposed model has the potential to achieve competitive performance with lower complexity. The adopted approach aims to support medical diagnosis with an intelligent, reliable, and more accessible tool for clinical settings with limited infrastructure, while maintaining computational efficiency.