EUVIMEDEUVIMED entdecken
Crossref

Pneumonia Screening and Prediction on Images Using a Convolutional Neural Network Deep Learning Model

Shuria Saaidin, Murizah Kassim, I Gede Pasek Suta Wijaya, Nur Amirah Kamaluddin

Pertanika Journal of Science and Technology · 30.10.2026 · 34 (5)

DOI: 10.47836/pjst.34.5.02 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Pertanika Journal of Science and Technology, online erschienen am 30.10.2026. Verfasst von Shuria Saaidin und 3 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Pneumonia remains a leading cause of morbidity and mortality worldwide, and its radiographic diagnosis depends on expert readers who are unavailable in many settings. This study reports a comparative evaluation of two ImageNet-pretrained convolutional architectures, VGG16 and Inception-v3, for binary pneumonia classification on chest radiographs. The publicly available Kaggle chest radiograph collection was used, restricted to the Normal and Pneumonia classes: 5,216 training images (1,341 normal, 3,875 pneumonia) and an independent test partition of 624 images (234 normal, 390 pneumonia). Each architecture was evaluated under two regimes: feature extraction with a frozen convolutional base and fine-tuning of the upper convolutional blocks. All performance figures are reported on the held-out test partition with 95% Wilson confidence intervals and are accompanied by specificity, balanced accuracy, and the Matthews correlation coefficient, as the test partition is imbalanced. The fine-tuned VGG16 model achieved an accuracy of 0.905 (0.880-0.926), sensitivity of 0.956, specificity of 0.821, F1-score of 0.927, and MCC of 0.797. The fine-tuned Inception-v3 model achieved an accuracy of 0.936 (0.914-0.953), sensitivity 0.951, specificity 0.910, F1-score 0.949 and MCC 0.863, at approximately 60% of the training time per epoch required by VGG16. Fine-tuning improved VGG16 by raising specificity at the cost of sensitivity but produced no measurable change in Inception-v3. These held-out figures are substantially below the near-perfect accuracies commonly reported for these architectures on this benchmark, a discrepancy attributable to the reporting of training-phase rather than test-phase performance.

Zitieren

APA

Saaidin, S., Kassim, M., Wijaya, I. G. P. S., & Kamaluddin, N. A. (2026). Pneumonia Screening and Prediction on Images Using a Convolutional Neural Network Deep Learning Model. Pertanika Journal of Science and Technology, 34(5). https://doi.org/10.47836/pjst.34.5.02

Vancouver

Saaidin S, Kassim M, Wijaya IGPS, Kamaluddin NA. Pneumonia Screening and Prediction on Images Using a Convolutional Neural Network Deep Learning Model. Pertanika Journal of Science and Technology. 2026;34(5). doi:10.47836/pjst.34.5.02

BibTeX
@article{Saaidin2026,
  author = {Saaidin, Shuria and Kassim, Murizah and Wijaya, I Gede Pasek Suta and Kamaluddin, Nur Amirah},
  title = {Pneumonia Screening and Prediction on Images Using a Convolutional Neural Network Deep Learning Model},
  journal = {Pertanika Journal of Science and Technology},
  year = {2026},
  volume = {34},
  number = {5},
  doi = {10.47836/pjst.34.5.02}
}
Kostenlos starten →