Skin Lesion Classification by Multi-View Filtered Transfer Learning

Jianxiao Bian,Song Zhang,Jianrui Zhang,Shaoqiang Wang,Jinchang Guo

doi:10.1109/access.2021.3076533

Abstract

Skin cancer is one of the most deadly cancer types with considerable number of patients. Image analysis has largely improved the automated diagnosis accuracy for malignant melanoma and other pigmented skin lesions, compared to unaided visual examination. Recent popular solution for automated skin lesion classification is using deep neural networks, trained from large amounts of professional annotated data, but that largely limits the model’s scalability. This paper exploits transfer learning for skin lesion classification task with the help of labeled data from another domain (source), and proposes a multi-view filtered transfer learning network to strongly represent discriminative information from different image views with reasonable weighing strategy. This method also evaluates the importance for each source images, which can learn useful knowledge with neglecting negative samples from source domain. The extensive skin lesion classification experiments demonstrate our method can effectively solve Melanoma and Seborrheic Keratosis classification tasks with outstanding extensibility, and the discussion of the major components also testifies the improvements of our proposed multi-view filtered transfer learning approach.

Highlights

Skin cancer expresses its severe harm to human health, where the malignancy leads to high fatality rate with frequently diagnosed around the world [1]
Research estimates that nonmelanoma skin cancer (NMSC), including basal cell carcinoma (BCC) and squamous cell carcinoma (SCC), affects more than 3 million Americans per year, and the overall incidence of BCC increased by 145% between 1976-1984 and 2000-2010, while the overall incidence of SCC increased 263% over that same period [2], [5]
ARLCNN [40] proposed an attention residual learning convolutional neural network for dermoscopy images, composing multiple ARL blocks, a global average pooling layer, and a classification layer; MCRes [41] developed a feasible multi-channel-resnet to assemble multiple residual neural networks for skin lesion analysis, pre-treating the training data with different methods; FusingDeep [42] combined intra-architecture and inter-architecture network fusion for skin lesion images, consisting of multiple sets of different CNN architectures that represent different feature abstraction levels; TransferFusion [14] proposed a novel fusion approach based on a three-level ensemble strategy that exploits multiple fine-tuned networks trained with dermoscopic images at various sizes

Summary

INTRODUCTION

Skin cancer expresses its severe harm to human health, where the malignancy leads to high fatality rate with frequently diagnosed around the world [1]. An effective domain adaptation method should further distill useful knowledge from the source domain and transfer the information into target domain These existing works [11]–[14] only reached poor performance compared to completely supervised skin cancer recognition models. B. CONTRIBUTION (1) We propose a Multi-view Filtered Transfer Learning (MFTL) method for cross-domain skin lesion classification to distill valuable source samples into adversarial domain adaptation and learn various informations from different image views. CONTRIBUTION (1) We propose a Multi-view Filtered Transfer Learning (MFTL) method for cross-domain skin lesion classification to distill valuable source samples into adversarial domain adaptation and learn various informations from different image views It can strengthen the representation capability for skin lesion images. The validating experiments achieve average AUC of 91.8% on ISIC 2017 dataset [15]

RELATED WORKS

FILTERED DOMAIN ADAPTATION

COMPARED WITH BASELINES

CONCLUSION