An improved hair removal algorithm for dermoscopy images

Dermoscopy is commonly used for diagnosing skin cancer in its early stages. However, hair structures in dermoscopy images can negatively affect diagnosis. A proposed algorithm based on conventional image processing methods removes hair structures from these images. Removing hair structures is crucial for constructing accurate computer-aided diagnosis systems for skin cancer. To eliminate hair structures from dermoscopy images, a hair removal algorithm based on conventional image processing methods has been proposed in this study. The proposed algorithm was compared to the existing hair removal algorithms in literature and tested on the ISIC2018 dataset using the AlexNet architecture. The effects on computer-aided systems were assessed by training on images with and without hair. Results show that the algorithm performs well in removing hairs compared to previous studies and improves classification performance. Upon comparison with other literature studies, the recommended algorithm has exhibited consistently high performance, usually ranking among the top two performers in the rank analysis. Additionally, the integration of the suggested algorithm has led to improvements in the performance metrics of the AlexNet architecture, with increases of 0.9% in accuracy, 1.4% in sensitivity, 0.6% in specificity, and 1.06 in F1 score. The performance of the suggested algorithm indicates its potential as a practical and effective tool in clinical settings.