303 A multi-task learning network for skin disease classification

In recent years, convolutional neural networks (CNNs), due to their powerful end-to-end feature learning capabilities, have been widely used in medical aided diagnosis. general, a dermatologist's diagnostic process first identifies the type of skin lesions, and then obtains diagnosis diseases according recognition lesions. However, most existing methods only focus on designing diverse directly identify diseases, while ignoring importance this paper, multi-task network is proposed for disease classification. network, addition classification task, multiple lesion tasks (e.g., cyst, plaque, macule) are added determine types thereby providing useful information Specifically, we concatenate features with original at top use concatenated final To verify effectiveness collected dataset Xiangya Hospital, Central South University, including 8713 clinical images five common (i.e., basal cell carcinoma, melanoma, nevus, squamous seborrheic keratosis). Preliminary experimental results show that performs better than methods.