@article {10.3844/jcssp.2026.1866.1880,
article_type = {journal},
title = {Polar Disentangled Non-Local Convolutional Neural Network for Skin Cancer Classification Using Dermoscopy Images},
author = {Thirugnanam, Usha and Joseph, Nalini},
volume = {22},
number = {6},
year = {2026},
month = {Jun},
pages = {1866-1880},
doi = {10.3844/jcssp.2026.1866.1880},
url = {https://thescipub.com/abstract/jcssp.2026.1866.1880},
abstract = {Skin cancer classification is the process of identifying and categorizing various types of skin lesions into cancerous and non-cancerous. This process is performed to achieve accurate detection, which minimizes the risk of severe health complications. However, classifying the severity of skin lesion images is a difficult task because of subtle and overlapping visual patterns, which could lead to inaccurate performance. To address this issue, this research proposes a Polar Disentangled Non-Local Convolutional Neural Network (PDNLCNN) to classify the skin images accurately by using dermoscopy images. In a conventional CNN, PDNL is incorporated to enhance the network’s ability to capture global contextual relationships effectively. This enables the model to capture fine-grained lesion information and structural patterns effectively, which are essential for accurate classification. Furthermore, the Fuzzy C-Means (FCM) clustering algorithm is employed to segment skin cancer, which manages uncertainty by assigning varying degrees of membership. This improves the boundary detection of the skin lesions and enables better accuracy in the segmentation process. Hence, the proposed PDNLCNN achieves superior classification accuracy of 99.21% on the HAM10000 dataset when compared with the existing methods, such as Deep CNN (DCNN).},
journal = {Journal of Computer Science},
publisher = {Science Publications}
}