EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Hawk-Swarmception Segmentation Network (HSCS-Net): Enhanced Liver Tumor Segmentation with Receptive Field Optimization and Clinical Data-Guided Feature Selection via PSO and GWO

Prachi Chhabra, Sukhendra Singh, Tathagat Banerjee

Artificial Intelligence in Applied Sciences · 2026

DOI: 10.69882/adba.ai.2026013 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Artificial Intelligence in Applied Sciences, 2026. Verfasst von Prachi Chhabra und 2 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The detection and treatment of cancerous tumors in the liver are considered one of the most challenging health issues for patients globally. There is a need for fully automated systems that can precisely detect and segment the tumorous regions in medical images. This work aims to develop an automated system which utilizes a hybrid- deep- learning framework with fusion multi-scale inception feature extraction called HSCS-NET (Hawk-Swarm Ception Segmentation Network). In the proposed model, the encoder comprises of Hawk Gating with SE (squeeze-and-excitation) attention while decoders consist of adaptive attention skip fusion which comprises of Swarm ception Residual ASPP bridges. These components allow the model to recover important details of the boundaries of the tumors irrespective to their shape, size, and the tissue contrast in CT scans.In order to improve segmentation, the HSCS-NET framework is equipped with a hybrid optimization module based on PSO (Particle Swarm Optimization) and GWO (Grey Wolf Optimization) for dynamic feature selection and reliable convergence. The model was evaluated against the 3DIRCADb1 Liver Tumor Segmentation Challenge (LiTS) dataset and significantly outperformed all other models achieving a Dice coefficient value of 0.98, Accuracy of 0.9891, and Precision of 0.9901.This marks a substantial improvement over prior models such as Christ et al.’s CNN (Dice: 0.823), Wu et al.’s Fuzzy C-means + GC (Dice: 0.83), Muhammad et al.’s ResNet (Dice: 0.87, Accuracy: 0.945, Precision: 0.93), and Kaur et al.’s PSO-PSP-Net (Accuracy: 0.9754, Precision: 0.9632).The HSCS-NET architecture is mathematically grounded, modularly extensible, and validated through rigorous cross-validation and confidence refinement. With its high segmentation performance, clinical reliability, and computational efficiency, HSCS-NET stands as a superior advancement in automated liver cancer diagnostics, reducing clinician workload and improving patient prognoses through precision imaging.

Zitieren

APA

Chhabra, P., Singh, S., & Banerjee, T. (2026). Hawk-Swarmception Segmentation Network (HSCS-Net): Enhanced Liver Tumor Segmentation with Receptive Field Optimization and Clinical Data-Guided Feature Selection via PSO and GWO. Artificial Intelligence in Applied Sciences. https://doi.org/10.69882/adba.ai.2026013

Vancouver

Chhabra P, Singh S, Banerjee T. Hawk-Swarmception Segmentation Network (HSCS-Net): Enhanced Liver Tumor Segmentation with Receptive Field Optimization and Clinical Data-Guided Feature Selection via PSO and GWO. Artificial Intelligence in Applied Sciences. 2026. doi:10.69882/adba.ai.2026013

BibTeX
@article{Chhabra2026,
  author = {Chhabra, Prachi and Singh, Sukhendra and Banerjee, Tathagat},
  title = {Hawk-Swarmception Segmentation Network (HSCS-Net): Enhanced Liver Tumor Segmentation with Receptive Field Optimization and Clinical Data-Guided Feature Selection via PSO and GWO},
  journal = {Artificial Intelligence in Applied Sciences},
  year = {2026},
  doi = {10.69882/adba.ai.2026013}
}
Kostenlos starten →