EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

An Optimized Self-Supervised Hybrid CNN–Transformer Model Using Metaheuristic Search for Robust Mammographic Breast Cancer Detection

Mohamed Abdulhasan Hussein

Journal of Artificial Intelligence in Medical Applications · 2026

DOI: 10.65511/jaima.v2i1.1796 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Artificial Intelligence in Medical Applications, 2026. Verfasst von Mohamed Abdulhasan Hussein. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

One of the most important difficulties in AI-based applications for healthcare is early diagnosis of breast cancer based on mammography images. Conventional models need vast volumes of expert-labeled data and their effectiveness is highly dependent on the choice of structure and proper training settings. The objective of this project was to establish an intelligent and efficient framework for diagnosis of breast cancer from mammography images using convolutional neural networks (CNNs) integrated with vision transformers and self-supervised learning and Metaheuristic Optimisation approaches. The suggested a hybrid model which combines convolutional neural networks to local feature extraction with vision transformers for global contextual connection analysis in medical images. Self-supervised learning was used in the first training phase to lessen dependence on annotated data using unlabelled photos. The Particle Swarm (PSO) and Ant Colony (ACO) algorithms were used to adapt the hyperparameters such as the learning rate, the patch size, the number of transformer layers, and the self-supervised learning parameters in order to automatically optimise the model performance. The model was assessed using the mammography image datasets DDSM and CBIS-DDSM. The findings indicated that the suggested model performed better than conventional models and other deep learning models, reaching classification accuracy of 98.1%, sensitivity of 97.6%, specificity of 97.9%, F1 value of 97.7%, and area under the ROC curve (AUC) of 0.98. The ablation investigation further indicated that the performance and generalisability could be greatly increased by using the combination of self-supervised learning and metaheuristic optimisation approaches.

Zitieren

APA

Hussein, M. A. (2026). An Optimized Self-Supervised Hybrid CNN–Transformer Model Using Metaheuristic Search for Robust Mammographic Breast Cancer Detection. Journal of Artificial Intelligence in Medical Applications. https://doi.org/10.65511/jaima.v2i1.1796

Vancouver

Hussein MA. An Optimized Self-Supervised Hybrid CNN–Transformer Model Using Metaheuristic Search for Robust Mammographic Breast Cancer Detection. Journal of Artificial Intelligence in Medical Applications. 2026. doi:10.65511/jaima.v2i1.1796

BibTeX
@article{Hussein2026,
  author = {Hussein, Mohamed Abdulhasan},
  title = {An Optimized Self-Supervised Hybrid CNN–Transformer Model Using Metaheuristic Search for Robust Mammographic Breast Cancer Detection},
  journal = {Journal of Artificial Intelligence in Medical Applications},
  year = {2026},
  doi = {10.65511/jaima.v2i1.1796}
}
Kostenlos starten →