A mobile AI-enabled platform for screening human skin diseases
Abstract
Computer-aided systems for skin lesion diagnosis is a growing area of research. Researchers have
recently shown an increasing interest in developing computer-aided diagnosis systems. This
project aims to review, synthesize and evaluate the quality of evidence for the diagnostic accuracy
of computer-aided systems. The experimentation is carried out on two publicly available datasets,
Dermnet and Dermquest. The dataset is partitioned into training data and validation data for each
category of skin disease. The train data consists of 80% of the dataset while the validation data
consists of 20%. In the proposed pre-processing, we apply lesion segmentation and augmentation.
This involves increasing the amount of data by adding slightly modified copies of already existing
data (geometry- and intensity-based). The algorithm used for training is the sequential CNN model
and the model is finally deployed on the developed Android application. The achieved accuracy
on the training and validation is 87% and 85% without data augmentation. After augmenting the
data, the accuracy of the training and validation data increases to 90.25% and 89.01% respectively.
The trained model provides better classification outcomes on the different datasets, leading to a
better recognition tool to assist dermatologists. With higher quality and a larger quantity of data,
it will be viable to use state-of-the-art models to enable the use of CAD in the field of dermatology.