ISIC Challenge 2020

先前的皮肤图像数据集尚未解决从同一患者的多个皮肤病变获得的患者级别信息。尽管人工智能分类算法在检查单个图像的对照研究中已达到专家级的性能，但实际上，皮肤科医生从同一患者的多个病变中全面判断。构建本文所述的2020 siim-isic黑色素瘤分类挑战数据集以解决先前挑战与临床实践之间的这种差异，为数据集中的每个图像提供允许来自同一患者的病变被映射到彼此的标识符。临床医生经常使用此患者水平的上下文信息来诊断黑色素瘤，并且在排除许多非典型痣患者的假阳性方面特别有用。该数据集代表来自三大洲的2,056名患者，平均每位患者16个病变，包括33,126个皮肤镜图像和584个组织病理学证实的黑色素瘤，与良性黑色素瘤模拟物相比。

Previous skin image datasets have failed to address patient-level information derived from multiple skin lesions of the same patient. While artificial intelligence classification algorithms have achieved expert-level performance in controlled studies examining single images, dermatologists actually make comprehensive diagnoses based on multiple lesions from the same patient. The 2020 SIIM-ISIC Melanoma Classification Challenge dataset described in this paper was constructed to address this discrepancy between prior challenges and clinical practice, providing each image in the dataset with identifiers that allow lesions from the same patient to be mapped to one another. Clinicians often use this patient-level contextual information to diagnose melanoma, and it is particularly useful for excluding false positives in patients with numerous atypical nevi. This dataset comprises 33,126 dermoscopic images from 2,056 patients across three continents, with an average of 16 lesions per patient, and includes 584 histopathologically confirmed melanomas as compared to benign melanoma mimics.

搜集汇总

背景与挑战

背景概述

ISIC Challenge 2020数据集旨在弥补先前皮肤图像数据集中患者级别信息的不足，通过提供来自2,056名患者的33,126个皮肤镜图像和584个组织病理学证实的黑色素瘤，支持基于多病变上下文的临床诊断。该数据集由昆士兰大学等机构于2020年发布，专注于黑色素瘤分类挑战。