New Compact Deep Learning Model for Skin Cancer Recognition

Deep learning neural networks have made significant progress in image analysis and have been used for skin cancer recognition. Early detection and proper treatments for malignant skin cancer cases are vital to ensure high survival rate in patients. We present a novel deep learning based convolutional neural network (CNN) model for generating compatible models on mobile platforms such as Android and iOS. The proposed model was tested on the grand challenge PHDB melanoma dataset. The best performing proposed model excels in the following ways: (1) it outperforms the baseline model in terms of accuracy by 1%; (2) it consists of 60% fewer parameters compared to the base model and thereby it is more efficient on mobile platforms. Furthermore, the model is more compact and retains high accuracy without the need to be downsized; (3) in conjunction with advanced regularization techniques such as dropout and data augmentation, the proposed CNN model excelled when implemented on state-of-the-art frameworks such as Keras and TensorFlow. Additionally, we were able to successfully deploy it on the iOS and Android mobile systems. The proposed model could also be lucrative towards other datasets for image classification on mobile platform.