Residual Network based Aggregation Model for Skin Lesion Classification
This work addresses skin lesion diagnosis, a critical medical imaging problem, but appears incremental as it combines existing techniques (residual networks and Fisher vectors) without claiming major breakthroughs.
The authors tackled skin lesion classification by proposing an aggregation algorithm that combines residual networks for local feature extraction with Fisher vector encoding for image-level representation, achieving results on the ISIC2018 challenge's disease classification task.
We recognize that the skin lesion diagnosis is an essential and challenging sub-task in Image classification, in which the Fisher vector (FV) encoding algorithm and deep convolutional neural network (DCNN) are two of the most successful techniques. Since the joint use of FV and DCNN has demonstrated proven success, the joint techniques could have discriminatory power on skin lesion diagnosis as well. To this hypothesis, we propose the aggregation algorithm for skin lesion diagnosis that utilize the residual network to extract the local features and the Fisher vector method to aggregate the local features to image-level representation. We applied our algorithm on the International Skin Imaging Collaboration 2018 (ISIC2018) challenge and only focus on the third task, i.e., the disease classification.