Multimodal skin lesion classification using deep learning
Problem
Prior automated skin lesion classifiers used only a single macroscopic image, limiting diagnostic performance in real-world clinical scenarios.
Why it matters
More accurate automated skin lesion diagnosis could reduce misdiagnosis of melanoma and other lesions.
What the AI did
A multimodal deep-learning classifier (CNN) combines multiple imaging modalities with patient metadata to improve automated skin lesion classification, outperforming a single-image baseline.
Results
Binary melanoma detection AUC 0.866 vs 0.784 baseline; multiclass mAP 0.729 vs 0.598
Impact
Improves automated skin lesion diagnosis over single-image baselines in benchmark evaluation.
Record
| Venue | Experimental Dermatology |
|---|---|
| Published | 2018 |
| Work type | article |
| Authors | Jordan Yap, William Yolland, Philipp Tschandl |
| Citations | 323 (OpenAlex, at import time) |
| References | 32 |
| Model or method | CNN (multimodal; specific architecture not named) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1111/exd.13777 |
| Open access | hybrid |
| Keywords | deep learning, CNN, skin lesion, multimodal, melanoma, dermatoscopy |
Importance rating
real but narrower advance
Automated skin lesion diagnosis including melanoma detection has clinical value, and multimodal input improves over single-image baselines, but the abstract presents a benchmark improvement on 2917 cases without stating broader scale or impact.