SavedAdd your journal

Journals / Mühendislik Bilimleri ve Araştırmaları Dergisi / 2026 / Cilt: 8 Sayı: 2

Comparative Analysis and Adaptation Strategies of Vision Foundation Models in Dermoscopic Image Segmentation

Dermoskopik Görüntü Segmentasyonunda Görsel Temel Modellerin Karşılaştırmalı Analizi ve Adaptasyon Stratejileri

Kenan Akyol(DICLE UNIVERSITY)0009-0001-2533-255X
Mesut Şeker(DİCLE ÜNİVERSİTESİ, MÜHENDİSLİK FAKÜLTESİ, ELEKTRİK-ELEKTRONİK MÜHENDİSLİĞİ BÖLÜMÜ)0000-0001-9245-6790
Pages
207–220
Publication Date
2026-10-27
🏛️Bibliographic metadata and full text access for this article are provided via DergiPark.
DergiPark Record ↗
✨

AI Summary & Key Takeaways

AI Insight
⚡ TL;DR(Too Long; Didn't Read)

This adapter-based framework is planned for integration into clinical decision-support systems and other medical imaging modalities.

📌 3 Key Takeaways

🎯
Objective & Hypothesis:Accurate segmentation of skin lesions is of critical clinical importance for early melanoma diagnosis; however, the direct application of large foundation models like the Segment Anything Model (SAM) to medical images characterized by low contrast and fuzzy boundaries remains insufficient due to substantial domain gaps.
📊
Key Findings & Data:In current study, a parameter-efficient fine-tuning strategy is presented by integrating lightweight adapter modules into a frozen Vision Transformer backbone.
💡
Conclusion & Impact:This adapter-based framework is planned for integration into clinical decision-support systems and other medical imaging modalities.
🔬Design:Modeling & Simulation
📍Scope:Comprehensive Sample
⚙️Variables:Deri Lezyonu Segmentasyonu · Segment Anything Model · SAM-Adapter

Abstract

Accurate segmentation of skin lesions is of critical clinical importance for early melanoma diagnosis; however, the direct application of large foundation models like the Segment Anything Model (SAM) to medical images characterized by low contrast and fuzzy boundaries remains insufficient due to substantial domain gaps. In current study, a parameter-efficient fine-tuning strategy is presented by integrating lightweight adapter modules into a frozen Vision Transformer backbone. The proposed SAM-Adapter approach was comparatively evaluated against the original SAM and MedSAM models on the publicly available ISIC2018 and HAM10000 dermoscopic datasets. Optimized with AdamW, the model achieved the highest Dice (0.9603) and IoU (0.9241) scores along with the lowest HD95 (22.57) value on the HAM10000 dataset. Validated as statistically significant via the Wilcoxon test, this approach minimized computational overhead while demonstrating superior accuracy in boundary delineation. This adapter-based framework is planned for integration into clinical decision-support systems and other medical imaging modalities.

Özet

Deri lezyonlarının doğru segmentasyonu, melanomun erken tanısında kritik öneme sahipken; Segment Anything Model (SAM) gibi büyük temel modellerin düşük kontrastlı ve belirsiz sınırlı tıbbi görüntülere doğrudan uygulanması yapısal alan farklılıkları nedeniyle yetersiz kalmaktadır. Bu çalışmada, dondurulmuş bir Vision Transformer omurgasına hafif adaptör modüllerinin entegre edildiği parametre verimli bir ince ayar stratejisi sunulmaktadır. Önerilen SAM-Adapter yöntemi; ISIC2018 ve HAM10000 açık erişimli dermoskopik veri setleri üzerinde orijinal SAM ve MedSAM modelleriyle karşılaştırmalı olarak değerlendirilmiştir. AdamW optimizasyonuyla eğitilen model; HAM10000 veri setinde en yüksek Dice (0.9603) ve IoU (0.9241) skorlarına ulaşırken, en düşük HD95 (22.57) değerini elde etmiştir. Wilcoxon testi ile istatistiksel anlamlılığı doğrulanan bu yaklaşım, hesaplama yükünü en aza indirerek zorlu lezyon sınırlarının belirlenmesinde üstün başarı göstermiştir. Geliştirilen adaptör tabanlı çerçevenin klinik karar destek sistemlerine ve farklı tıbbi görüntüleme modalitelerine entegrasyonu hedeflenmektedir.

🔖

Cite This Article

Akyol, K., Şeker, M. (2026). Comparative Analysis and Adaptation Strategies of Vision Foundation Models in Dermoscopic Image Segmentation. Mühendislik Bilimleri ve Araştırmaları Dergisi, Cilt: 8 Sayı: 2, 207–220. https://doi.org/10.46387/bjesr.2018519

📄 Full Text PDF Viewer

📄

PDF görüntüleyici hazırlanıyor...

References (19)

  1. F. Bray et al., “Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries,” CA. Cancer J. Clin., vol. 74, no. 3, pp. 229–263, May 2024.
  2. H. Naseri and A. A. Safaei, “Diagnosis and prognosis of melanoma from dermoscopy images using machine learning and deep learning: a systematic literature review,” BMC Cancer, vol. 25, no. 1, p. 75, 2025.
  3. L. Mevorach, A. Farcomeni, G. Pellacani, and C. Cantisani, “A Comparison of Skin Lesions’ Diagnoses Between AI-Based Image Classification, an Expert Dermatologist, and a Non-Expert,” 2025.
  4. M. Yang and N. I. R. Ruhaiyem, “MCM-UNet: Mamba convolutional mixing network for skin lesion image segmentation,” Biomed. Signal Process. Control, vol. 112, p. 108791, 2026.
  5. O. Ronneberger, P. Fischer, and T. Brox, “U-Net: Convolutional Networks for Biomedical Image Segmentation,” May 2015, [Online]. Available: http://arxiv.org/abs/1505.04597
  6. F. Milletari, N. Navab, and S. Ahmadi, “V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation,” in 2016 Fourth International Conference on 3D Vision (3DV), 2016, pp. 565–571.
  7. A. Dosovitskiy, L. Beyer, A. Kolesnikov, and D. Weissenborn, “An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale,” in ICLR 2021, 2021. [Online]. Available: https://arxiv.org/abs/2010.11929
  8. Y. Wang et al., “Attention-enhanced U-Net based network for cancerous tissue segmentation,” Biomed. Signal Process. Control, vol. 106, p. 107728, 2025.
  9. F. Li et al., “Semantic-SAM: Segment and Recognize Anything at Any Granularity,” arXiv Prepr. arXiv2307.04767, pp. 1–14, 2023, [Online]. Available: https://arxiv.org/abs/2307.04767
  10. A. Esteva et al., “Dermatologist-level classification of skin cancer with deep neural networks,” Nature, vol. 542, no. 7639, pp. 115–118, 2017.
  11. J. Ma, Y. He, F. Li, L. Han, C. You, and B. Wang, “Segment anything in medical images,” Nat. Commun., vol. 15, no. 1, p. 654, Jan. 2024.
  12. Z. Su et al., “Adapting SAM to Histopathology Images for Tumor Bud Segmentation in Colorectal Cancer.,” Proc. SPIE--the Int. Soc. Opt. Eng., vol. 12933, Feb. 2024.
  13. F. Binzagr and M. Hariri, “Foundation-Model-Driven Skin Lesion Segmentation and Classification Using SAM-Adapters and Vision Transformers,” 2026.
  14. M. Hu, Y. Li, and X. Yang, “SkinSAM: Empowering Skin Cancer Segmentation with Segment Anything Model,” in CVPR, 2023. [Online]. Available: https://arxiv.org/abs/2304.13973
  15. J. Wu et al., “Medical SAM adapter: Adapting segment anything model for medical image segmentation,” Med. Image Anal., vol. 102, p. 103547, May 2025.
  16. T. Chen et al., “SAM Fails to Segment Anything? -- SAM-Adapter: Adapting SAM in Underperformed Scenes: Camouflage, Shadow, Medical Image Segmentation, and More,” Apr. 2023.
  17. T. Chen et al., “SAM-Adapter: Adapting Segment Anything in Underperformed Scenes,” in 2023 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), 2023, pp. 3359–3367.
  18. S. Gül, B. M. Aydın, D. Akgün, R. Öztaş Kara, and G. Çetinel, “Skin Lesion Segmentation with Semantic SAM: Pros and Cons,” Selcuk Univ. J. Eng. Sci. Vol 23, No 3 Selcuk Univ. J. Eng. Sci., Dec. 2024, [Online]. Available: https://sujes.selcuk.edu.tr/sujes/article/view/660
  19. Ö. F. Söylemez, “SkinAttn-Net: a multi-level attention-based network for skin lesion segmentation,” Sci. Rep., vol. 16, no. 1, p. 3700, 2025.