BME Frontiers / 2022 / Article / Tab 3

Research Article

A Low-Cost High-Performance Data Augmentation for Deep Learning-Based Skin Lesion Classification

Table 3

The performance of searched method on ISIC 2017 dataset (refer to https://challenge.isic-archive.com/leaderboards/2017 (Task 3: Lesion Diagnosis)) and derm_7pt dataset.

Team/authorsUse external dataUse ensemble modelsAvg. AUCBACCSensitivity (MEL, SK)Specificity (MEL, SK)Dataset

Xie et al. [29]YesYes0.9380.727, 0.8440.915, 0.945ISIC 2017
Zhang et al. [30]YesNo0.9170.878 (mean)0.867 (mean)
Matsunaga et al. [31]YesYes0.9110.8310.735, 0.9780.851, 0.773
González et al. [32]NoYes0.9100.8830.103, 0.1780.998, 0.998
Menegola et al. [33]YesYes0.9080.8440.547, 0.3560.950, 0.990
Yu et al. [34]YesYes0.897
Yang et al. [35]NoNo0.8860.8090.350,0.5560.96, 0.976
Our approach (EfficientNet b1)NoNo, ,
Kawahara et al. [16]YesYes0.8960.6040.604 (mean)0.91 (mean)Derm7pt
Tudor et al. [36]YesYes0.6380.638 (mean)0.926 (mean)
Rodrigues et al. [37]YesYes0.620.4080.408 (mean)0.710 (mean)
Alzahrani et al. [38]NoNo0.6380.638 (mean)0.702 (mean)
Our approach (EfficientNet b0)NoNo (mean) (mean)