Classification of psoriasis and
eczema based on Convolutional Neural Network
الباحث الأول:
Marwa Ali Alhamdani
الباحثين الآخرين:
Asaad Noori Hashim
المجلة:
2022 Iraqi International Conference on Communication
and Information Technologies (I
تاريخ النشر:
13 يونيو، 2023
مختصر البحث:
Dermatology is one of the most difficult terrains to diagnose and predict due to its complexity. In the field of dermatology, many tests are often required to identify what skin disease a patient may suffer from. Thus, a system that can diagnose ski…
Dermatology is one of the most difficult terrains to diagnose and predict due to its complexity. In the field of dermatology, many tests are often required to identify what skin disease a patient may suffer from. Thus, a system that can diagnose skin diseases without such limitations is required. Medical technology has shown tremendous progress in distinguishing psoriasis from eczema faster and more accurately; however, the cost of this diagnosis is still limited and expensive. That is why, the current research proposes a convolution neural network (CNN) approach to develop and test a system that makes a decisive distinction between psoriasis and eczema, using a database of 5564 images that are collected from (derm Net, Kaggle) websites and then reclassified by a dermatologist, to train a CNN, which is uses to extract the medium and high-level features from the input data. The proposed structure of the proposed classification network is composed of four layers, three of which are the convolution layers while the fourth layer is the dense layer. Moreover, the CNN has proven its efficiency in learning relevant features from images. The proposed system has achieved an accuracy average of 96.13% in the classification after testing several parameters that have a significant impact on the classification accuracy including (the number of epochs, number of layers, filter size, and size of the input images) as well as testing the effect of hyperparameters such as activation function, dropout batch size optimizer in addition to testing the effect of image optimization. This accuracy is excellent in comparison to similar works.
