[Paper Review] A Novel Multi-task Deep Learning Model for Skin Lesion Segmentation and Classification

Xulei Yang, Zeng Zeng+4 more|arXiv (Cornell University)|Mar 3, 2017

Cutaneous Melanoma Detection and Management8 references78 citations

TL;DR

The paper proposes a multi-task deep neural network for simultaneous skin lesion segmentation and two binary classifications, showing competitive ISIC 2017 results.

ABSTRACT

In this study, a multi-task deep neural network is proposed for skin lesion analysis. The proposed multi-task learning model solves different tasks (e.g., lesion segmentation and two independent binary lesion classifications) at the same time by exploiting commonalities and differences across tasks. This results in improved learning efficiency and potential prediction accuracy for the task-specific models, when compared to training the individual models separately. The proposed multi-task deep learning model is trained and evaluated on the dermoscopic image sets from the International Skin Imaging Collaboration (ISIC) 2017 Challenge - Skin Lesion Analysis towards Melanoma Detection, which consists of 2000 training samples and 150 evaluation samples. The experimental results show that the proposed multi-task deep learning model achieves promising performances on skin lesion segmentation and classification. The average value of Jaccard index for lesion segmentation is 0.724, while the average values of area under the receiver operating characteristic curve (AUC) on two individual lesion classifications are 0.880 and 0.972, respectively.

Motivation & Objective

• Motivate multi-task learning to exploit shared representations across segmentation and classification tasks in skin lesion analysis.
• Develop a unified deep neural network that performs segmentation and two binary lesion classifications concurrently.
• Evaluate the approach on ISIC 2017 data to assess learning efficiency and potential accuracy gains over independent models.

Proposed method

• Propose a multi-task deep neural network that shares representations across tasks (segmentation and two binary classifications).
• Train and evaluate on ISIC 2017 Dermoscopic images (2000 training, 150 evaluation).
• Use commonalities and differences across tasks to improve task-specific performance.
• Report performance using Jaccard index for segmentation and AUC for each classification task.

Experimental results

Research questions

• RQ1 Can a single multi-task network improve segmentation and classification performance compared to independently trained models?
• RQ2 To what extent does shared representation learning enhance learning efficiency and accuracy for skin lesion analysis?

Key findings

• Average Jaccard index for segmentation: 0.724.
• AUC for first classification: 0.880.
• AUC for second classification: 0.972.
• Model evaluated on ISIC 2017 dataset with 2000 training and 150 testing samples.
• Multi-task learning is reported to offer improved learning efficiency and potential accuracy gains over separate models.