|
KMID : 0390320230330020039
|
|
Chungbuk Medical Journal
2023 Volume.33 No. 2 p.39 ~ p.46
|
|
|
Classification Of Skin Lesions Based On Inception-Resnet-V2
|
|
Park Jae-Sung
Oh Jeong-Wook
Lee Tae-Soo
|
|
|
Abstract
|
|
|
|
Skin cancer is an increasing worldwide health problem. The recent studies have reported that excessive ultraviolet rays exposure must have been a main cause in developing skin cancer. The best solution for skin cancer is a timely diagnosis of skin lesions as the five-year survival rate for melanoma patients is 97% when diagnosed at the early stage. Considering the difficulty of dermatologists for accurate diagnosis of skin cancer, we need to develop an automated efficient system for skin lesions classification. We utilized Inception-resnet-v2 model for transfer learning with 10,015 skin lesion images from Harverd Dataverse (HAM10000). As a result, the model based on transfer learning achieved an accuracy of 90.41% for 2 classes; melanocytic nevi vs. the other lesions. Also, we found an accuracy of 93.16% for melanocytic nevi vs. melanoma. This model has the potential to support dermatologists in decision making about skin lesions.
|
|
|
KEYWORD
|
|
|
Transfer Learning, Pre-trained Neural Network, Deep Learning, AI, CNN
|
|
|
FullTexts / Linksout information
|
|
|
|
|
|
Listed journal information
|
|
|
|