Akademik Veri Yönetim Sistemi
9th International Symposium on Multidisciplinary Studies and Innovative Technologies, ISMSIT 2025, Ankara, Türkiye, 14 - 16 Kasım 2025, (Tam Metin Bildiri)
Dysplastic nevi are special types of dermatological nevus that resemble melanoma. They are difficult to diagnose and require support for histopathological evaluation to confirm whether they are progressing to malignancy. Due to the invasive, impractical, and not cost-effective nature of the histopathological evaluation process for the patient, the necessity arises to apply effective methods that are an alternative to the histopathological evaluation approach. In order to create this alternative solution, in this research, the unique spectral identification information that hyperspectral images have at the pixel level due to their spectral bands is benefited. Nowadays, it is being tried for distinction between dysplastic nevus and melanoma by understanding this pixel information with automatic classification methods. Within the scope of this study, it was aimed to find the most effective classification method by comparing current classification models such as SMESC (Spatial Mapping Expansion with Spectral Compression), DBSSAN (Dual-Branch Spectral-Spatial Attention Network), MSSTT (Multiscale Super Token Transformer), VAN (Visual Attention Network), ResNet50, ResNet101, C3D (Convolutional 3D), ViT-b16 (Vision Transformer), EfficientNet-b3, and Inception-v3 on a publicly available dermoscopy hyper-spectral dataset. According to the experimental results of Overall Accuracy (OA), Average Accuracy (AA), Kappa, Precision, Recall, and F1_score, it was observed that MSSTT was the most successful method. However, as a result of the evaluations made, the need for the use of hyperspectral dermoscopy images in diagnosis and the development of more effective classification models for the use of these images in diagnosis is emphasized in this study. Furthermore, this study highlights the need for a large, publicly available hyperspectral dataset for the discrimination of dermatological lesions.