Related Experiment Video
Updated: Oct 17, 2025
09:37
Combining Reflectance Confocal Microscopy with Optical Coherence Tomography for Noninvasive Diagnosis of Skin Cancers via Image Acquisition
Published on: August 18, 2022
2.6K
Acral melanoma detection using dermoscopic images and convolutional neural networks
Qaiser Abbas1, Farheen Ramzan2, Muhammad Usman Ghani2
1Department of Computer Science, University of Engineering and Technology, 54890, Lahore, Pakistan. mqaiser617@gmail.com.
Visual Computing for Industry, Biomedicine, and Art
|October 7, 2021
Summary
This study introduces a deep learning model for classifying acral melanoma (AM), a rare skin cancer. The model achieved over 90% accuracy, aiding dermatologists in early AM diagnosis.
Area of Science:
- Dermatology and Artificial Intelligence
- Computational Pathology
- Medical Imaging Analysis
Background:
- Acral melanoma (AM) is a rare, lethal skin cancer with diagnostic challenges due to subtle differences from benign lesions.
- Current research primarily focuses on binary melanoma classification, with limited work on melanoma subtypes.
- Dermoscopic imaging is crucial for expert diagnosis, but subtle distinctions remain difficult.
Purpose of the Study:
- To investigate the effectiveness of deep learning and dermoscopy in classifying melanoma subtypes, specifically acral melanoma (AM).
- To develop and evaluate a novel deep learning model for automated skin cancer classification, including AM detection.
- To compare the performance of a custom deep learning model against transfer learning approaches using established architectures.
Main Methods:
- Utilized a dermoscopic image dataset from Yonsei University Health System, South Korea.
- Applied image processing and data augmentation techniques to enhance the automated system for AM detection.
- Developed a seven-layered deep convolutional network trained from scratch and compared it with fine-tuned AlexNet and ResNet-18 models using transfer learning.
Main Results:
- The custom deep learning model achieved over 90% accuracy in classifying acral melanoma and benign nevus.
- Transfer learning approaches, using modified AlexNet and ResNet-18, yielded an average accuracy of nearly 97%.
- The proposed system demonstrated robust performance, comparable to state-of-the-art methods in skin cancer classification.
Conclusions:
- The developed deep learning model effectively classifies skin cancer subtypes, including acral melanoma.
- The system shows potential as a clinical decision support tool for dermatologists, facilitating early AM diagnosis.
- Combining dermoscopy with advanced deep learning techniques offers a promising avenue for improving skin cancer diagnostics.
