Deep Learning Applied to COVID-19 Detection in Chest X-Rays: A Transfer Learning Approach

This study investigates the application of deep learning to the classification of COVID-19 in chest X-ray images using transfer learning. The MobileNetV2 architecture, pretrained on the ImageNet dataset, was used to classify radiographic images into three categories: COVID-19, normal, and pneumonia. The dataset contains 6,432 images distributed across the three classes. The experimental procedure included resizing the images to 224×224 pixels, normalizing pixel values, data augmentation, and two-stage training. Initially, the pretrained convolutional base was kept frozen while a new classification layer was trained. Next, some of the network's final layers were unfrozen for fine-tuning, using a reduced learning rate. In the final evaluation reported for the test set, the model achieved an accuracy of 94.72% and a macro F1-score of 94.54%. For the COVID-19 class, precision of 97.67%, sensitivity of 96.55%, and an F1-score of 97.11% were obtained. The results indicate that the approach showed promising performance in classifying the images analyzed. However, its application as a tool to support clinical diagnosis requires further validation using independent datasets and evaluation in real clinical settings. Keywords: deep learning; COVID-19; chest X-rays; MobileNetV2; transfer learning; image classification.