Application of YOLO Models for Assisted Tumor Diagnosis: A Computer Vision-Based Approach

Resumen: This study emphasizes the application of deep learning techniques for tumor detection in medical images using YOLOv11. By leveraging convolutional neural networks (CNNs) pre-trained on large-scale datasets, the model fine-tunes its parameters to address the specific challenge of tumor identification. Training and validation are analyzed through loss and metric graphs. The losses associated with bounding box adjustment, classification, and coordinates showed a progressive decrease, indicating continuous improvement in model accuracy during both training and validation. Key metrics such as precision, recall, and mAP were evaluated. The model achieved an accuracy exceeding 95%, while recall increased from 75% to 90%, demonstrating robust capability in identifying true positives. The confusion matrix analysis revealed 287 correctly classified tumors, with 24 false negatives and 36 false positives, suggesting areas for improvement. The results confirm the effectiveness of transfer learning in medical image analysis tasks. The developed model performs accurate detections, making it a computationally efficient and viable solution for applications in resource-constrained environments. This study lays a solid foundation for the development of computer vision-based diagnostic tools that could enhance early tumor detection and improve clinical outcomes in vulnerable populations.