EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Lesion-Aware Vision Transformer and Mamba Fusion for Lightweight Dermoscopic Skin Cancer Classification

Tanzila Saba

Journal of Intelligent Decision Making and Information Science · 09.09.2026 · 3 (18s) · 719-735

DOI: 10.17762/jidmis.v3.5754 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Intelligent Decision Making and Information Science, online erschienen am 09.09.2026. Verfasst von Tanzila Saba. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Skin cancer is one of the most destructive malignancies of the current decade and the fifth most common cancer. Automated dermoscopic skin lesion diagnosis is a foundational task in computer-aided diagnosis (CAD). Early clinical detection of melanoma drastically increases the 5-year patient survival rate from below 15% in advanced Stage IV to over 98% in localized Stage IA. To address these challenges while maintaining parameter efficiency suitable for edge-device deployment, this work proposes the Lesion-Aware Fusion MambaFormer (LAF-MambaFormer), a streamlined, three-stage hybrid architecture engineered for robust binary classification (Benign vs. Malignant). With a compact budget of exactly 15.85M trainable parameters, the model combines local inductive bias, global multi-head self-attention, selective state-space sequence modeling, and lesion-aware boundary refinement. Stage 1 leverages a pre-trained ConvNeXt-Nano Multi-Scale Feature Encoder (MSFE) to extract hierarchical feature maps. Stage 2 unites Transformer self-attention (TGCM) and Mamba Selective State-Space (MLDM) sequence modeling with cross-modal Hadamard interaction. Stage 3 enriches features via the Lesion-Aware Fusion Module (LAFM) and a Boundary Detail Enhancement Module (BDEM) using a discrete Laplacian edge extraction cascade. Results thus obtained confirm that the proposed model provides reliable classification with minimal false negatives, offering practical advantages for clinical dermatology.

Zitieren

APA

Saba, T. (2026). Lesion-Aware Vision Transformer and Mamba Fusion for Lightweight Dermoscopic Skin Cancer Classification. Journal of Intelligent Decision Making and Information Science, 3(18s), 719-735. https://doi.org/10.17762/jidmis.v3.5754

Vancouver

Saba T. Lesion-Aware Vision Transformer and Mamba Fusion for Lightweight Dermoscopic Skin Cancer Classification. Journal of Intelligent Decision Making and Information Science. 2026;3(18s):719-735. doi:10.17762/jidmis.v3.5754

BibTeX
@article{Saba2026,
  author = {Saba, Tanzila},
  title = {Lesion-Aware Vision Transformer and Mamba Fusion for Lightweight Dermoscopic Skin Cancer Classification},
  journal = {Journal of Intelligent Decision Making and Information Science},
  year = {2026},
  volume = {3},
  number = {18s},
  pages = {719--735},
  doi = {10.17762/jidmis.v3.5754}
}
Kostenlos starten →