AUTHOR=Ahmad Naveed , Shah Jamal Hussain , Khan Muhammad Attique , Baili Jamel , Ansari Ghulam Jillani , Tariq Usman , Kim Ye Jin , Cha Jae-Hyuk 

TITLE=A novel framework of multiclass skin lesion recognition from dermoscopic images using deep learning and explainable AI

JOURNAL=Frontiers in Oncology

VOLUME=Volume 13 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2023.1151257

DOI=10.3389/fonc.2023.1151257

ISSN=2234-943X

ABSTRACT=Skin cancer is a severe disease worldwide. Melanoma is one of the aggressive forms of skin cancer, and early diagnosis of melanoma can significantly reduce the human mortality rate. However, challenges such as lesion irregularities, low contrast lesions, color similarity in intra-class, redundant features, and imbalanced datasets are extremely challenging for improved recognition accuracy. This work presented a new framework for skin lesion recognition using data augmentation, deep learning, and explainable AI. In the proposed framework, data augmentation is performed at the initial step to increase the dataset size, and then two pre-trained deep learning models are employed. Both models have been fine-tuned and trained using deep transfer learning. The global average pooling layer is utilized by both models (Xception and ShuffleNet) for deep feature extraction. The analysis of this step shows that some important information is missing; therefore, we performed the fusion. After the fusion process, the computational time was increased; therefore, we developed an improved Butterfly Optimization Algorithm (BOA). Using this algorithm, only the best features are selected and classified using Machine Learning (ML) classifiers. In addition, a GradCAM based visualization is performed to analyze the important region in the image. Two publically available datasets – ISIC2018 and HAM10000 have been utilized and obtained improved accuracy of 99.3 and 91.5%, respectively. Comparing the proposed framework accuracy with state-of-the-art methods, the results reveal improved and less computational time.