Segmentation of Skin Lesions and their Attributes Using Multi-Scale Convolutional Neural Networks and Domain Specific Augmentations

Abstract

Computer-aided diagnosis systems for classification of different type of skin lesions have been an active field of research in recent decades. It has been shown that introducing lesions and their attributes masks into lesion classification pipeline can greatly improve the performance. In this paper, we propose a framework by incorporating transfer learning for segmenting lesions and their attributes based on the convolutional neural networks. The proposed framework is based on the encoderdecoder architecture which utilizes a variety of pre-trained networks in the encoding path and generates the prediction map by combining multi-scale information in decoding path using a pyramid pooling manner. To address the lack of training data and increase the proposed model generalization, an extensive set of novel domain-specific augmentation routines have been applied to simulate the real variations in dermoscopy images. Finally, by performing broad experiments on three different data sets obtained from International Skin Imaging Collaboration archive (ISIC2016, ISIC2017, and ISIC2018 challenges data sets), we show that the proposed method outperforms other state-of-the-art approaches for ISIC2016 and ISIC2017 segmentation task and achieved the 1 st rank on the leader-board of ISIC2018 attribute detection task.

Automated skin lesion segmentation using multi-scale feature extraction scheme and dual-attention mechanism

“…Even though these methods have shown satisfactory performance by enhancing the capability to extract local features, they fail to extract required global features for significantly higher performance. Recently by combining multiple backbones, Jahanifar et al [26] proposed an ensemble approach for the segmentation of skin lesions. But the ensemble of multiple models will increase the number of parameters, and such models require more run-time for network convergence.…”

Stain-Robust Mitotic Figure Detection for the Mitosis Domain Generalization Challenge

“…Using this model with pre-trained weights from ImageNet as a backbone allows the overall model to benefit from transfer-learning, by extracting better feature representations and gaining higher domain generalizability. The Jaccard loss function [9] is robust against the imbalanced population of positive and negative pixels in the segmentation dataset, and thus has been utilised to train the model.…”

NuClick: A Deep Learning Framework for Interactive Segmentation of Microscopy Images

“…In order to segment both small and large objects, the network must be able to capture features on various scales. Therefore, we incorporate multi-scale convolutional blocks [79] throughout the network (with specific design configurations related to the network level). Unlike other network designs (eg.…”