SkinAACN: An Efficient Skin Lesion Classification Based on Attention Augmented ConvNeXt with Hybrid Loss Function

The accurate diagnosis and treatment of skin lesions require precise identification. Traditional approaches rely on the expertise of dermatologists, creating a demand for more efficient methods. Recent advancements in deep learning have facilitated the development of intelligent systems for the detection and classification of dermoscopic images. However, existing models often struggle to selectively focus on relevant image regions, leading to reduced classification accuracy. This paper introduces an attention-augmented ConvNeXt network designed to address this limitation. The proposed model incorporates diverse attention mechanisms, including channel and spatial attention, enhancing its ability to focus on informative image segments. Furthermore, a hybrid loss function combining cross-entropy and triplet loss was utilized during training to improve feature embedding and class separation. Our experiments on the HAM10000 dataset show that our model outperforms the ConvNeXt baseline, with the Efficient Channel Attention (ECA) augmented model achieving the highest accuracy of 94.89