Scholarlink

Pneumonia continues to be one of the leading causes of morbidity and mortality globally, particularly affecting children under five years of age and older adults in regions with limited resources. Faced with challenges in timely diagnosis using chest x-rays, especially due to the shortage of radiologists and subjectivity in interpretation, artificial intelligence (AI) is presented as a key tool for strengthening diagnostic accuracy. In this study, a hybrid deep learning approach for automatic classification of chest images was proposed, evaluating different visual feature extractors (LeViT, MobileViT, DeiT, SwinTransformer, PiT, ViT) combined with classifiers such as SVM, Random Forest and K-NN. The results revealed an outstanding performance without hyperparameter adjustment (tuning), reaching maximum values of accuracy of 99%, precision of 99%, recall of 98% and F1-score of 99%, being the best performance obtained with the ViT-SVM combination. These findings highlight the potential of hybrid models in supporting automated medical diagnosis. Thus, the integration of AI in clinical contexts could significantly reduce diagnostic errors, improve health coverage in vulnerable areas and contribute to more efficient early detection.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Robust Model for Pneumonia Detection from X-Rays Using Machine Learning and Deep Learning Techniques

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • George Peña,
  • Wilfredo Ticona

摘要

Pneumonia continues to be one of the leading causes of morbidity and mortality globally, particularly affecting children under five years of age and older adults in regions with limited resources. Faced with challenges in timely diagnosis using chest x-rays, especially due to the shortage of radiologists and subjectivity in interpretation, artificial intelligence (AI) is presented as a key tool for strengthening diagnostic accuracy. In this study, a hybrid deep learning approach for automatic classification of chest images was proposed, evaluating different visual feature extractors (LeViT, MobileViT, DeiT, SwinTransformer, PiT, ViT) combined with classifiers such as SVM, Random Forest and K-NN. The results revealed an outstanding performance without hyperparameter adjustment (tuning), reaching maximum values of accuracy of 99%, precision of 99%, recall of 98% and F1-score of 99%, being the best performance obtained with the ViT-SVM combination. These findings highlight the potential of hybrid models in supporting automated medical diagnosis. Thus, the integration of AI in clinical contexts could significantly reduce diagnostic errors, improve health coverage in vulnerable areas and contribute to more efficient early detection.