Paper Key : IRJ************050
Author: Vishakha Shalikram Patil,Saurabh Mehta
Date Published: 06 Nov 2023
Abstract
Dermatology is the name for the area of medicine that deals with skin. Due to the intricacy, it is considered to be one of the clinical branch's hardest areas to diagnose. Skin cancer and dermatological illnesses are very challenging to visually diagnose in the early to middle stages of the disease. Additionally, the time it takes to diagnose a skin condition might differ from practitioner to practitioner and also be influenced by the practitioner's level of experience. If an illness, such as melanoma, is not treated in a timely manner, it may have very serious consequences. Therefore, a method that can identify skin illnesses without any of these limitations is required. Self-assessment may also prove to be the best option in areas where proper resources are few, such as in rural and congested areas. We suggest a machine learning-based, automated method for processing images to identify skin diseases. The skin disease image is initially subjected to several pre-processing methods used in image processing. Machine learning methods are used to identify disorders in second stage. Unwanted noise is removed from skin photos. Convolutional neural networks (CNNs) will be a key component of the system's overall synthesis. We are proposing three models on which we did training and further testing is done by test data set. Three distinct CNN models were designed and trained using a diverse dataset of dermatological images. The models were evaluated using a rigorous cross-validation technique to assess their performance with regard to accuracy. The findings shows that Model 1 achieved an accuracy of 80.5%, Model 2 reached 85%, and Model 3 exhibited an accuracy of 65% in the categorization of skin diseases. Our study highlights the potential of CNN-based models in automating the detection of skin diseases, with Model 2 demonstrating the highest accuracy among the three models. These results highlight deep learning's potential in the dermatological discipline and suggest that further research and even more accuracy can result from model design advancements and reliable skin disease detection systems. This suggested method will concentrate on obtaining more precise findings, and it may be applied globally without a lot of expensive equipment or resource needs.
DOI LINK : 10.56726/IRJMETS45747 https://www.doi.org/10.56726/IRJMETS45747