Skin Lesion Classification Using Pre-Trained DenseNet201 Deep Neural Network
S.P. Godlin Jasil, V. Ulagamuthalvi
Abstract
Skin cancer is the most common dangerous type of cancer now a days. Because of the heterogeneity in appearance of skin lesion the expert find difficulty to detect from dermoscopic images. We propose a deep neural network using DenseNet201 pretrained architecture that classifies the seven classes of dermoscopic images for early detection of skin cancer. The proposed network is trained by International Skin Imaging Collaboration as a part of ISIC 2018 challenge that consist of 2487 training images and 604 test images. Our result show 95% accuracy of training images and 77% accuracy of test image.
Topics & Concepts
Artificial intelligenceSkin cancerComputer scienceArtificial neural networkSkin lesionDeep learningPattern recognition (psychology)Deep neural networksContextual image classificationLesionNetwork architectureTest (biology)Image (mathematics)CancerComputer visionDermatologyMedicinePathologyPaleontologyBiologyComputer securityInternal medicineCutaneous Melanoma Detection and ManagementNonmelanoma Skin Cancer StudiesAI in cancer detection