EUVIMEDEUVIMED entdecken
Crossref

Automated Medical Image Segmentation Using Hybrid Transfer Learning Attention Network

Rathiya R, suja GP, Harsha BK, Prakash Kanchikere Mathad, Sk. Hedayath Basha, Kavitha Shanmugakani

International Journal of Medical Toxicology and Forensic Medicine · 19.09.2026

DOI: 10.22037/ijmtfm.v16.53261 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Medical Toxicology and Forensic Medicine, erschienen am 19.09.2026. Verfasst von Rathiya R und 5 weiteren Autor:innen.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

Diagnosis of brain tumors using magnetic resonance imaging (MRI) is essential in clinical diagnosis and treatment planning since it is accurate and allows for early diagnosis. However, there is a shortage of annotated medical data and inter-class similarity among tumor types, posing a challenge for automated classification systems. The proposed research is a Hybrid Transfer Learning Attention Network (HTLA-Net) for multi-class brain tumor classification, based on pre-trained convolutional neural networks and stage-by-stage fine-tuning.

Methods

The proposed architecture of HTLA-Net is based on a combination of backbone-specific pre-processing, feature extraction, and selective layer modification to expand learning while maintaining consistent generalization. The authors evaluated four state-of-the-art architectures, i.e., EfficientNetB0, ResNet50, MobileNetV2, and DenseNet121, on a publicly available four-class MRI dataset, comprising glioma, meningioma, pituitary macroadenoma, and healthy cases.

Results

The experiment shows that DenseNet121 achieved the best overall performance, with a test accuracy of 85.96, a macro F1-score of 0.8557, a specificity of 0.9532, and an ROC-AUC of 0.9597, attributable to its strong ability to distinguish between classes and its robust diagnostic performance.

Conclusion

The proposed hybrid training plan is an effective way to mitigate overfitting and improve calibration compared to simpler architectures. These findings affirm the suitability and efficiency of deep models that rely on transfer learning for reliable computer-based diagnosis of brain tumors in settings with limited medical practice and data.

Schlagwörter

brain tumor classificationmagnetic resonance imagingtransfer learningattention mechanismmulti-class classificationDenseNet121gliomameningiomapituitary macroadenomafine-tuning

Zitieren

APA

R, R., GP, S., BK, H., Mathad, P. K., Basha, S. H., & Shanmugakani, K. (2026). Automated Medical Image Segmentation Using Hybrid Transfer Learning Attention Network. International Journal of Medical Toxicology and Forensic Medicine. https://doi.org/10.22037/ijmtfm.v16.53261

Vancouver

R R, GP S, BK H, Mathad PK, Basha SH, Shanmugakani K. Automated Medical Image Segmentation Using Hybrid Transfer Learning Attention Network. International Journal of Medical Toxicology and Forensic Medicine. 2026. doi:10.22037/ijmtfm.v16.53261

BibTeX
@article{R2026,
  author = {R, Rathiya and GP, suja and BK, Harsha and Mathad, Prakash Kanchikere and Basha, Sk. Hedayath and Shanmugakani, Kavitha},
  title = {Automated Medical Image Segmentation Using Hybrid Transfer Learning Attention Network},
  journal = {International Journal of Medical Toxicology and Forensic Medicine},
  year = {2026},
  doi = {10.22037/ijmtfm.v16.53261}
}
Kostenlos starten →