Plataforma web inteligente para la identificación automática de tipos de leucemia usando Deep Learning

• Autores: Santillan Chota, Bruno Alejandro, Mendoza Rios, Alessandra del Pilar
• Formato: tesis de grado
• Fecha de Publicación: 2026
• Institución: Universidad Nacional De La Amazonía Peruana
• Repositorio: UNAPIquitos-Institucional
• Lenguaje: español
• OAI Identifier: oai:repositorio.unapiquitos.edu.pe:20.500.12737/12854
• Enlace del recurso: https://hdl.handle.net/20.500.12737/12854
• Nivel de acceso: acceso abierto
• Materia: Navegador web; Inteligencia Artificial; Diagnodtico; Leucemia; https://purl.org/pe-repo/ocde/ford#2.02.04

The present research aimed to implement a web platform based on Deep Learning algorithms to diagnose and predictively classify four main types of leukemia (Acute Lymphoblastic Leukemia, Acute Myeloid Leukemia, Chronic Myeloid Leukemia, and Chronic Lymphocytic Leukemia) in the Iquitos population. The methodology employed a pre-experimental design with a quantitative approach and predictive level, evaluating 67 digitized images of blood smears from patients with confirmed diagnosis by a hematology specialist from the Regional Hospital of Loreto. The platform was developed using deep convolutional neural networks implemented in Python with TensorFlow/Keras and React, processing microscopic images through adaptive preprocessing algorithms. Results demonstrated exceptional diagnostic performance with 98.51% accuracy (66/67 correct classifications), Cohen's Kappa coefficient of 0.9528 indicating almost perfect agreement with specialized clinical criteria, weighted average precision of 99.39%, sensitivity of 96.67%, and specificity of 99.39%. The platform exhibited satisfactory functional operability with inference latency of 0.78 seconds per image and system availability of 99.2%, meeting 100% of the evaluated technical criteria. It is concluded that the Deep Learning-based web platform constitutes a viable diagnostic tool for automation of complex hematological processes in hospital contexts with limited resources, achieving functional equivalence with specialized clinical judgment and exceeding international reference thresholds established by regulatory agencies for medical artificial intelligence systems.