Automate Skin Lesion Classification Using Deep Neural Network
| Main Author: | Nguyen, Thanh |
|---|---|
| Format: | info software Journal |
| Terbitan: |
, 2019
|
| Subjects: | |
| Online Access: |
https://zenodo.org/record/2656095 |
Daftar Isi:
- One of the most common form of cancer in the United States is skin cancer. There are three forms of skin cancer with melanoma being the deadliest but also the least common out of them. Early detection is critical due to the survival rate for early detection of melanoma however, with rates going down significantly for each stage with no treatment. Skin cancers are often diagnosed using a tool called dermatoscope. This study automates skin cancer diagnosis process by using convolutional neural network (CNN) to build a classifier to detect different types of skin lesions. The models were built using Pytorch pretrained models, AlexNet and ResNet. AlexNet has two main layers, a feature extraction layer composed of 5 convolutional layers, and a classifier layer composed of 3 fully connected (FC) layers with a dropout layer after every FC layer. All the models had high accuracy on the test set with the lowest was 73% and the highest was 82%. Although, there was limitation with the dataset, key features of the images used the study showed a classification model that had a good accuracy and good predicting power toward the different forms of cancer; creating a good baseline for future study in skin cancer classification.