A Low-Cost High-Performance Data Augmentation for Deep Learning-Based Skin Lesion Classification
Problem
Skin lesion classification needs high-performance, low-cost data augmentation so intelligent skin cancer screening devices can work in rural or underdeveloped communities.
Why it matters
Early skin cancer screening in low-resource settings affects diagnosis and survival for underserved populations.
What the AI did
A plug-and-play data augmentation strategy (search space 10^1) searches for the best augmentation method for a medical database, combined with deep learning baseline models.
Results
Best BACC of 0.853 on HAM10000, outperforming other published single-model, no-external-database models for ISIC 2018.
Impact
Improved benchmark classification performance and highlighted regions of interest for clinicians; no clinical deployment shown.
Record
| Venue | BME Frontiers |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Shuwei Shen, Mengjuan Xu, Fan Zhang, Pengfei Shao, Honghong Liu, Liang Xu, Chi Zhang, Peng Liu et al. |
| Citations | 61 (OpenAlex, at import time) |
| References | 44 |
| Model or method | EfficientNets (baseline) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.34133/2022/9765307 |
| Open access | diamond |
| Keywords | data augmentation, skin lesion, deep learning, EfficientNet, HAM10000, low-resource |
Importance rating
real but narrower advance
Aims at low-cost skin cancer screening for rural/underdeveloped communities; benchmark gains on public datasets, but abstract gives no population scale or deployment evidence.