Hybrid Detection Techniques for Skin Cancer Images
MetadataShow full item record
According to W.H.O, skin cancer is one of the most common types of human malignancy in medical sector. Application of deep learning is continuously being used to determine the accuracy of detecting different medical problems more effectively. A lot of new techniques have been discovered to fast forward the procedure with having highest percentage of accuracy. In this research work, we have proposed a model to detect skin cancer more effectively using image processing with convolutional neural network, a part of deep learning concept under machine learning. The dataset contains almost 3000+ images of the patients having skin diseases classified into two classes, malignant and benign. We have introduced CNN along with its seven different architectures to find the accuracy of the images of skin cancer and performed a comparative analysis to find out the best architecture that suits this type of problem. We ran ResNet50, VGG16, InceptionV3, VGG19, Xception, MobileNetV2, MobileNet architectures to find out the best model for that suits our type of dataset. Our model could be made even better if we changed the parameters, i.e. increasing the value of epochs, lowering batch size, changing the value of dropout etc. which would take longer period of times obviously. In our model, MobilenetV2 gave us the lowest accuracy with having almost 54.545% accuracy. The best architecture that we found for our dataset was the Xception. This architecture gave us about almost 85.303% accuracy for our dataset. Due to low power computer and other factors like, dataset image quality, image pre-processing, getting less accuracy can arise. © 2020 IEEE.