Classification of skin cancer in darker complexions : A case study

Skin cancer detection models using AI have attained exceptional standards, but frequently exhibit bias due to the insufficient representation of darker skin tones in datasets, which results in diagnostic inconsistency. Therefore, this study aims to improve the performance of an AI model for the classification of skin cancer in darker complexions with the help of transfer learning and data augmentation. Utilizing the Fitzpatrick17K dataset, which includes annotations of skin types, we trained and fine-tuned a neural network to maximize the classification accuracy for skin cancer in darker complexions. The results of our study is that detection of skin cancer in diverse data improved. The findings from our study is that transfer learning based on a model pre-trained on a generic dataset can improve model performance in darker complexions.