Comprehensive Summary
Al-Waisy et al. present Skin-DeepNet, a tool used to identify and classify skin cancer lesions from early dermoscopy images. While dermoscopy imaging has enhanced the diagnosis of skin cancer from manual diagnosis, it has difficulty recognizing early-stage lesions accurately. Skin-DeepNet uses three steps: image pre-processing, image segmentation, and image extraction to collect the most relevant information from each image. The system was then tested on the ISIC 2019 and HAM1000 datasets, which utilized images from eight categories of skin lesions. Results showed that Skin-DeepNet performed with outstanding proficiency, having a 99.65% accuracy rate on the ISIC 2019 dataset and a 100% accuracy on the HAM1000 dataset, in addition to the near-perfect precision, recall, F1-score, and AUC values.
Outcomes and Implications
Skin cancer is a major health issue around the world. To improve survival rates among patients, early and accurate diagnosis of skin lesions is necessary. Currently, the method used for diagnosis is visual examination by a dermatologist, which can often be unreliable, as it has an accuracy rate of 60%. The authors developed the Skin-DeepNet system, which performed with near-perfect results in classifying dermoscopic images to aid in the diagnosis of early skin cancer. By providing accurate results, this system can significantly reduce misdiagnosis and improve physician confidence and patient outcomes. While these results are quite promising, the authors emphasize the need for large-scale clinical trials to better understand its real-world effectiveness before implementing it into clinical practice.