Diagnosing Skin Diseases Across Diverse Tones using AI

Aquil, Akasha and Saeed, Faisal and Elmitwally, Nouh (2024) Diagnosing Skin Diseases Across Diverse Tones using AI. In: The 4th International Conference of Advanced Computing and Informatics, 16 - 17 December 2024, Birmingham City University. (In Press)

[thumbnail of ICACIN_2024_Camera_Ready_Paper_88.pdf]

Text
ICACIN_2024_Camera_Ready_Paper_88.pdf - Accepted Version
Restricted to Repository staff only
Download (213kB) | Request a copy

Abstract

Skin diseases in melanin rich skin often present diagnostic challenges, as their detection is more difficult due to the characteristics of darker skin tones, leading to misdiagnosis or delayed treatment. This disparity affects millions of individuals in the Black community. Developing accurate and re-liable AI-based automated tools is crucial to address these challenges. In this project, we applied three machine learning models and a data mining ap-proach to accurately predict various skin conditions using dermoscopic im-ages. Employing the CRISP-DM methodology, we implemented three tech-niques: Convolutional Neural Networks (CNNs), known for their strength in image recognition; Support Vector Machines (SVMs), effective in distin-guishing subtle differences in skin conditions; and Decision Trees, a simpler yet interpretable model. These were applied to predict skin disease types us-ing the HAM10000 image dataset. To manage data imbalance, SMOTE and resampling techniques were employed. Additionally, PCA and fine-tuning were used to optimize the models. Accuracy is selected as evaluation met-rics. Among the models, SVM with the RBF kernel achieved the highest ac-curacy of 88.48%, followed by Decision Trees with 82.48% and CNN-ResNet50 with 77.28%. Comparisons with other HAM10000 dataset predic-tion methods demonstrated superior performance for the proposed approach, particularly with the SVM-RBF model. These results suggest that our meth-ods for skin disease detection have significant potential to benefit healthcare by offering more accurate and reliable diagnostic tools.

Item Type:	Conference or Workshop Item (Paper)
Dates:	Date Event 17 December 2024 Accepted
Subjects:	CAH11 - computing > CAH11-01 - computing > CAH11-01-01 - computer science
Divisions:	Faculty of Computing, Engineering and the Built Environment > College of Computing
Depositing User:	Nouh Elmitwally
Date Deposited:	13 Feb 2025 12:41
Last Modified:	13 Feb 2025 12:45
URI:	https://www.open-access.bcu.ac.uk/id/eprint/16137