Assisted deep learning framework for multi-class skin lesion classification considering a binary classification support
Problem
Classifying dermoscopy images into seven lesion classes is difficult, and multi-class classification accuracy needed improvement.
Why it matters
Better automated skin lesion classification supports more reliable skin cancer diagnosis.
What the AI did
A dual-task CNN framework used the more reliable healthy/diseased binary classification confidences to tune the multi-class confidence values, implemented with GoogLeNet Inception-v3.
Results
Balanced multi-class accuracy remarkably raised by 7% via embedding binary classification outcomes.
Impact
Experiments demonstrated that binary-assisted confidence tuning improved multi-class dermoscopy classification accuracy.
Record
| Venue | Biomedical Signal Processing and Control |
|---|---|
| Published | 2020 |
| Work type | article |
| Authors | Balázs Harangi, Ágnes Baran, András Hajdú |
| Citations | 79 (OpenAlex, at import time) |
| References | 19 |
| Model or method | GoogLeNet Inception-v3 (dual-branch framework) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1016/j.bspc.2020.102041 |
| Open access | hybrid |
| Keywords | CNN, dermoscopy, skin lesion, multi-class classification, Inception-v3, binary classification |
Importance rating
real but narrower advance
Automated skin lesion classification supports cancer diagnosis, but abstract only demonstrates a 7% balanced-accuracy gain on a benchmark; no scale or clinical impact stated.