EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

U-Net Based Framework for Automatic Skin Lesion Segmentation

Bhanuprathap S, Rashmi S, Hemachandran K, Rakshitha S, Pramodha M, Channabasava Chola

Procedia Computer Science · 2026 · 283 · 2392-2401

DOI: 10.1016/j.procs.2026.06.309 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Procedia Computer Science, 2026. Verfasst von Bhanuprathap S und 5 weiteren Autor:innen. Die Arbeit nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Convolutional neural network based deep learning approaches have proven highly effective in interpreting medical images. U-net has emerged as a popular architecture for skin lesion segmentation and perform effectively across diverse dermoscopic datasets. Despite its success the traditional u-net with uniform downsampling layers in encoder and simple stacked convolution, may struggle to capture feature at multiple scales. Multiscale feature capture is crucial given the complexity and variability of skin lesions. To address these limitations, we developed an improved u-net framework. This framework incorporates Grad-Cam to provide interpretable visualizations while preserving fine spatial details. We evaluated the framework on the HAM10000 dataset using cross-validation. The results showed an average accuracy of 96.87%, precision of 95.63% and recall of 93.55%. Additional segmentation metrics confirmed the model’s reliability, including a dice score of 94.58%, IoU of 89.72% and a ROC AUC of 99.56%. Visualization analyses with Grad-Cam revealed the network consistently focused on clinically significant regions, particularly along lesion edges and in areas with complex textures. These results suggest that the framework delivers both accurate lesion segmentation and understandable results, positioning it as a promising approach for automated dermatological assessments.

Zitieren

APA

S, B., S, R., K, H., S, R., M, P., & Chola, C. (2026). U-Net Based Framework for Automatic Skin Lesion Segmentation. Procedia Computer Science, 283, 2392-2401. https://doi.org/10.1016/j.procs.2026.06.309

Vancouver

S B, S R, K H, S R, M P, Chola C. U-Net Based Framework for Automatic Skin Lesion Segmentation. Procedia Computer Science. 2026;283:2392-2401. doi:10.1016/j.procs.2026.06.309

BibTeX
@article{S2026,
  author = {S, Bhanuprathap and S, Rashmi and K, Hemachandran and S, Rakshitha and M, Pramodha and Chola, Channabasava},
  title = {U-Net Based Framework for Automatic Skin Lesion Segmentation},
  journal = {Procedia Computer Science},
  year = {2026},
  volume = {283},
  pages = {2392--2401},
  doi = {10.1016/j.procs.2026.06.309}
}
Kostenlos starten →