Abstract:
"This Research presents an image classification teledermatology app that can identify common skin
diseases in Sri Lanka. The app is particularly relevant due to the increased awareness and attention
towards skin diseases following the outbreak of Monkeypox during the 2021 pandemic. This
project has chosen 7 prevalent skin diseases among Sri Lankans: Acne, Chickenpox, Eczema,
Monkeypox, Psoriasis, Ringworm, and Rosacea. And this suggest to provide Sri Lankan-based
medical advice, taking into account the traditional and Ayurveda treatments commonly used in the
country. Skin health has a direct impact on mental health, and with the increasing number of
patients suffering from depression, reducing skin diseases can help alleviate mental health issues.
However, some people may be hesitant to seek medical treatment for skin conditions due to the
appearance of lesions and privacy concerns. To address this, the author proposes a teledermatology
concept that allows users to record and maintain a self-journal application, which stores images
and videos of their skin disease lesions to track their progress.
For this research project the author created a new dataset of 5600 original images by combining
existing datasets and collecting Sri Lankan patient’s data. And the author implemented an
Augmentation model to augment 25 images from one image using the Skimage library. And the
Author evaluated the image classification model using ResNet, EfficientNet and MobileNet with
Transfer Learning and CNN custom model. Finally Author selected the EfficientNet Transfer
learning model for feature extraction due to the high accuracy and evaluation matrix results. And
researcher combined feature extraction model with simple neural network model for classification
with merge model. And the author verified that the EfficientNet model has not been used in
existing skin disease classification models and it received 98% of overall accuracy."
