Skin Lesion Classification through Sequential Triple Attention DenseNet: Diverse Utilization of The Combination of Attention Modules
Shafiullah Shafin, Anwar Hossain Efat, S. M. Mahedy Hasan, Nahrin Jannat, Mahjabin Oishe, Mostarina Mitu, Mahib Uz Zaman
Abstract
Skin lesions refer to a broad term encompassing any skin abnormality, while skin diseases are conditions impacting the skin’s structure and function. Skin cancer, a specific skin disease, arises from uncontrolled abnormal skin cell growth. Early detection of lesions is crucial, but a challenge lies in precisely identifying regions within images, especially after accounting for dominant image classes. This article introduces an innovative approach, Sequential Triple Attention DenseNet (STAD), utilizing Convolutional Neural Networks (CNNs), DenseNet201, three attention mechanisms, and ensemble learning (EL). STAD focuses on salient features, pinpointing significant regions for classification. EL enhances performance and generalization. Various STAD model combinations contribute diverse feature maps, yielding improved outcomes. Empirical evaluation on the HAM1000 dataset achieves an impressive 99.18% accuracy, surpassing existing methods, highlighting its superiority. These findings hold promise for more efficient skin lesion recognition and skin cancer prevention techniques.