EUVIMEDEUVIMED entdecken
Qualitative StudieOpen Access · CC BY 4.0

Qualitative Study on Impact of EfficientNet-Based Deep Transfer Learning Model for Pneumonia Detection with Explainable Artificial Intelligence using Chest Radiographs

USMAN, Opeyemi Lateef, ADEUSI, Oluwasegun Opeoluwa, KAREEM, Morufat Adebola, OWOADE, Akeem Ayoade, MUNIYANDI, Ravie Chandren

Federal University Gusau Faculty of Education Journal · 25.03.2026 · 3 (1) · 45-56

DOI: 10.64348/zije.2026344 ↗

Kurz gefasst

Qualitative Studie in der Zeitschrift Federal University Gusau Faculty of Education Journal, online erschienen am 25.03.2026. Verfasst von USMAN und 9 weiteren Autor:innen. Die Arbeit nennt 27 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Pneumonia continues to constitute a significant global public health burden, with a disproportionate impact on children, older adults, and immunocompromised populations. Although chest radiography is routinely employed for diagnosis, manual image interpretation is frequently constrained by subjectivity, heavy clinical workloads, and inter-observer variability. This study proposes an EfficientNet-based deep transfer learning framework for automated pneumonia detection from chest radiographs, incorporating an explainable artificial intelligence (XAI) mechanism to improve model interpretability. The model was trained and validated using the widely adopted Kaggle pediatric chest X-ray dataset under multiple train–validation–test partitioning schemes (80:10:10, 70:15:15, 60:20:20, and 75:15:10). Model performance was evaluated using accuracy, precision, recall, F1-score, and the area under the receiver operating characteristic curve (ROC-AUC). In addition, Gradient-weighted Class Activation Mapping (Grad-CAM) was applied to generate visual explanations of the model’s predictions. The proposed approach demonstrated consistently strong performance across all data splits, with the 60:20:20 configuration yielding the highest accuracy (90.5%), recall (99.0%), and F1-score (94.9%). Despite the high accuracy and recall, the ROC-AUC scores were moderate (0.68–0.73), suggesting residual challenges in effectively distinguishing pneumonia cases from normal images. Grad-CAM visualizations indicated that the model predominantly focused on clinically meaningful pulmonary regions, thereby enhancing decision transparency and fostering clinical confidence. Overall, the findings indicate that the integration of EfficientNet, transfer learning, and explainable AI provides a robust and interpretable solution for pneumonia detection from chest radiographs. Future research will aim to validate the model on external datasets and investigate more advanced interpretability techniques to further strengthen diagnostic reliability.

Zitieren

APA

USMAN, Lateef, O., ADEUSI, Opeoluwa, O., KAREEM, Adebola, M., OWOADE, Ayoade, A., MUNIYANDI, & Chandren, R. (2026). Qualitative Study on Impact of EfficientNet-Based Deep Transfer Learning Model for Pneumonia Detection with Explainable Artificial Intelligence using Chest Radiographs. Federal University Gusau Faculty of Education Journal, 3(1), 45-56. https://doi.org/10.64348/zije.2026344

Vancouver

USMAN, Lateef O, ADEUSI, Opeoluwa O, KAREEM, Adebola M, et al. Qualitative Study on Impact of EfficientNet-Based Deep Transfer Learning Model for Pneumonia Detection with Explainable Artificial Intelligence using Chest Radiographs. Federal University Gusau Faculty of Education Journal. 2026;3(1):45-56. doi:10.64348/zije.2026344

BibTeX
@article{USMAN2026,
  author = {USMAN and Lateef, Opeyemi and ADEUSI and Opeoluwa, Oluwasegun and KAREEM and Adebola, Morufat and OWOADE and Ayoade, Akeem and MUNIYANDI and Chandren, Ravie},
  title = {Qualitative Study on Impact of EfficientNet-Based Deep Transfer Learning Model for Pneumonia Detection with Explainable Artificial Intelligence using Chest Radiographs},
  journal = {Federal University Gusau Faculty of Education Journal},
  year = {2026},
  volume = {3},
  number = {1},
  pages = {45--56},
  doi = {10.64348/zije.2026344}
}
Kostenlos starten →