A deep learning-based dual-branch framework for automated skin lesion segmentation and classification via dermoscopic Images
Problem
Limited access to dermatological expertise creates a need for automated skin lesion segmentation and classification from dermoscopic images.
Why it matters
Early skin disease detection significantly improves patient survival rates.
What the AI did
A dual-branch deep learning framework simultaneously segments skin lesions (modified EfficientNet-B7 encoder with ASPP, transformer blocks, attention gates) and classifies them by fusing DenseNet-121 features with mask-derived morphological characteristics.
Results
Highest Dice score 0.9568 on HAM10000; strong consistent segmentation across five benchmark datasets
Impact
Strong benchmark segmentation performance on five public datasets, including HAM10000.
Record
| Venue | Scientific Reports |
|---|---|
| Published | 2025 |
| Work type | article |
| Authors | Hamza Abu Owida, Ibrahim Abd El‐Fattah, Suhaila Farhan Ahmad Abuowaida, Nawaf Alshdaifat, Hamza A. Mashagba, Azlan B. Abd Aziz, Alaa AlZoubi, Samia Larguech et al. |
| Citations | 23 (OpenAlex, at import time) |
| References | 51 |
| Model or method | EfficientNet-B7 encoder, DenseNet-121, transformer blocks |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41598-025-21783-z |
| Open access | gold |
| Keywords | deep learning, skin lesion, dermoscopy, segmentation, classification, EfficientNet-B7 |
Importance rating
real but narrower advance
Early skin disease detection improves survival and dermatology access is limited, but abstract only demonstrates benchmark performance, no clinical scale.