Early skin disease detection is crucial for both effective treatment and the prevention of spreading to others. Neglected Tropical Skin Diseases (skin-NTDs) primarily affect low-income and developing countries, receiving inadequate attention and resources in comparison to other health issues. In this study, an automated system has been developed for detecting five skin-NTDs, capable of recognizing the diseases from raw lesion images without requiring any pre-processing. A dataset was created in the initial phase of the study since no publicly available benchmarks are available. The EfficientNet family of pre-trained models was utilized to train the classifier, and the EfficientNet-B3 was selected based on the experimental results. Additionally, the proposed work has developed a Grad-CAM based visualization technique to identify the most influential regions within images for the classification of specific diseases. The proposed model exhibited an overall classification accuracy of 91.53% on the test data. The proposed work will offer benefits to frontline medical staff and local residents in low-income countries.
BT - 2023 5th International Conference on Advancements in Computing (ICAC) DO - 10.1109/icac60630.2023.10417550 LA - Eng M3 - Conference Abstract N2 -Early skin disease detection is crucial for both effective treatment and the prevention of spreading to others. Neglected Tropical Skin Diseases (skin-NTDs) primarily affect low-income and developing countries, receiving inadequate attention and resources in comparison to other health issues. In this study, an automated system has been developed for detecting five skin-NTDs, capable of recognizing the diseases from raw lesion images without requiring any pre-processing. A dataset was created in the initial phase of the study since no publicly available benchmarks are available. The EfficientNet family of pre-trained models was utilized to train the classifier, and the EfficientNet-B3 was selected based on the experimental results. Additionally, the proposed work has developed a Grad-CAM based visualization technique to identify the most influential regions within images for the classification of specific diseases. The proposed model exhibited an overall classification accuracy of 91.53% on the test data. The proposed work will offer benefits to frontline medical staff and local residents in low-income countries.
PB - IEEE PY - 2023 T2 - 2023 5th International Conference on Advancements in Computing (ICAC) TI - Detection and Visualization of Neglected Tropical Skin Diseases Using EfficientNet and Grad-CAM ER -