A novel melanoma prediction model for imbalanced data using optimized SqueezeNet by bald eagle search optimization

• Published in: Computers in biology and medicine 2021-09, Vol.136, p.104712-104712, Article 104712
• Main Authors: Sayed, Gehad Ismail, Soliman, Mona M., Hassanien, Aboul Ella
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial intelligence; Artificial neural networks; Bald eagle search optimization; Classification; Computer architecture; Datasets; Deep learning; Dermoscopy; Humans; Hyperparameter optimization; Imbalanced data; Lesions; Melanoma; Melanoma - diagnostic imaging; Neural networks; Neural Networks, Computer; Optimization; Prediction models; Skin cancer; Skin diseases; Skin lesions; Skin Neoplasms - diagnostic imaging; Transfer learning
• ISSN: 0010-4825; 1879-0534
• DOI: 10.1016/j.compbiomed.2021.104712

Skin lesion classification plays a crucial role in diagnosing various gene and related local medical cases in the field of dermoscopy. In this paper, a new model for the classification of skin lesions as either normal or melanoma is presented. The proposed melanoma prediction model was evaluated on a large publicly available dataset called ISIC 2020. The main challenge of this dataset is severe class imbalance. This paper proposes an approach to overcome this problem using a random over-sampling method followed by data augmentation. Moreover, a new hybrid version of a convolutional neural network architecture and bald eagle search (BES) optimization is proposed. The BES algorithm is used to find the optimal values of the hyperparameters of a SqueezeNet architecture. The proposed melanoma skin cancer prediction model obtained an overall accuracy of 98.37%, specificity of 96.47%, sensitivity of 100%, f-score of 98.40%, and area under the curve of 99%. The experimental results showed the robustness and efficiency of the proposed model compared with VGG19, GoogleNet, and ResNet50. Additionally, the results showed that the proposed model was very competitive compared with the state of the art. [Display omitted] •New hybrid version of convolutional neural network architecture and BEO is proposed.•New melanoma skin cancer prediction model is introduced.•The proposed melanoma prediction model is evaluated on ISIC 2020 and ISIC 2019.•The proposed model outperforms other state-of-the-art.