[논문 리뷰] Automatic Skin Lesion Analysis using Large-scale Dermoscopy Images and Deep Residual Networks

Lei Bi, Jin‐Man Kim외 2명|arXiv (Cornell University)|2017. 03. 12.

Cutaneous Melanoma Detection and Management참고 문헌 8인용 수 173

한 줄 요약

논문은 자동 피부 병변 분석을 위해 대규모 dermoscopy 이미지에서 강건한 특징을 학습하기 위해 deep residual networks (ResNets)를 사용하는 것을 제안하며, 특히 ISIC 2017 챌린지에서 melanoma 탐지에 초점을 둡니다.

ABSTRACT

Malignant melanoma has one of the most rapidly increasing incidences in the world and has a considerable mortality rate. Early diagnosis is particularly important since melanoma can be cured with prompt excision. Dermoscopy images play an important role in the non-invasive early detection of melanoma [1]. However, melanoma detection using human vision alone can be subjective, inaccurate and poorly reproducible even among experienced dermatologists. This is attributed to the challenges in interpreting images with diverse characteristics including lesions of varying sizes and shapes, lesions that may have fuzzy boundaries, different skin colors and the presence of hair [2]. Therefore, the automatic analysis of dermoscopy images is a valuable aid for clinical decision making and for image-based diagnosis to identify diseases such as melanoma [1-4]. Deep residual networks (ResNets) has achieved state-of-the-art results in image classification and detection related problems [5-8]. In this ISIC 2017 skin lesion analysis challenge [9], we propose to exploit the deep ResNets for robust visual features learning and representations.

연구 동기 및 목표

• 주관성과 인간 해석의 변동성으로 인한 조기 멜라노마 진단 보조를 위한 자동 dermoscopy 분석의 필요성 제시.
• 대규모 dermoscopy 데이터셋을 활용하여 강건한 시각 특징을 학습.
• skin lesion 분류 및 탐지의 이미지 기반 성능을 개선하기 위해 deep residual networks 적용.

제안 방법

• Utilize deep residual networks (ResNets) for robust feature learning from dermoscopy images.
• Train on large-scale dermoscopy datasets to capture diverse lesion characteristics.
• Aim to improve classification/detection performance for melanoma within the ISIC 2017 challenge framework.

실험 결과

연구 질문

• RQ1 Can deep residual networks learn robust representations from large-scale dermoscopy images for skin lesion analysis?
• RQ2 Do ResNets improve melanoma-related classification/detection performance in dermoscopy compared to prior methods?
• RQ3 How well does the approach perform within the ISIC 2017 Challenge setting?

주요 결과

• The approach demonstrates the use of ResNets to learn robust features for skin lesion analysis from large-scale dermoscopy images.
• The work targets melanoma detection through automatic analysis in dermoscopy images.
• The method is positioned for evaluation in the ISIC 2017 Challenge.
• The paper reports progress toward leveraging deep learning for non-invasive early detection of melanoma.