LRprosy, also known as Hansen’s Disease is a chronic curable infection that seldom causes skin lesion and nerve damage. The World Health organization reported 127558 new leprosy cases detected globally in 2020. National Leprosy Eradication Program (NLEP) initiated by India, is the largest leprosy eradication program of the world. Despite this, 53.64% of leprosy cases (120,000 to 130,000 per year) are in India. Although this disease is curable at the later stages, an early diagnosis eradicates the complications incurred by nerve involvement. Existing research predicts leprosy on basis of electronic health records which require complex contact based techniques for procuration. Image based predictors lack credibility as it is hard to determine what features are being taken into consideration for prediction. This paper proposes an optimal AXI-CNN architecture for leprosy prediction called LeprosyNet. Deep Learning models are seldom black box in nature, in order to understand the area of interest in the images, explainable AI techniques like Activation Layer Visualization, Occlusion Sensitivity and Grad-CAM are used. The proposed model is compared with famous state-of-art architectures AlexNet and ResNet. Evaluation of the model has been undertaken using a ROC curve and confusion matrix. Accuracy obtained on LeprosyNet is 98%.
BT - 2023 International Conference on Sustainable Computing and Data Communication Systems (ICSCDS) DO - 10.1109/icscds56580.2023.10104958 LA - Eng N2 -LRprosy, also known as Hansen’s Disease is a chronic curable infection that seldom causes skin lesion and nerve damage. The World Health organization reported 127558 new leprosy cases detected globally in 2020. National Leprosy Eradication Program (NLEP) initiated by India, is the largest leprosy eradication program of the world. Despite this, 53.64% of leprosy cases (120,000 to 130,000 per year) are in India. Although this disease is curable at the later stages, an early diagnosis eradicates the complications incurred by nerve involvement. Existing research predicts leprosy on basis of electronic health records which require complex contact based techniques for procuration. Image based predictors lack credibility as it is hard to determine what features are being taken into consideration for prediction. This paper proposes an optimal AXI-CNN architecture for leprosy prediction called LeprosyNet. Deep Learning models are seldom black box in nature, in order to understand the area of interest in the images, explainable AI techniques like Activation Layer Visualization, Occlusion Sensitivity and Grad-CAM are used. The proposed model is compared with famous state-of-art architectures AlexNet and ResNet. Evaluation of the model has been undertaken using a ROC curve and confusion matrix. Accuracy obtained on LeprosyNet is 98%.
PB - IEEE PY - 2023 T2 - 2023 International Conference on Sustainable Computing and Data Communication Systems (ICSCDS) TI - Leprosy Diagnosis using Explainable Artificial Intelligence Techniques ER -