Carlos Alberto Arce-Lopera - Department of Information Technology, Faculty of Engineering, Universidad Icesi, Cali, Colombia
Javier Diaz-Cely - Department of Information Technology, Faculty of Engineering, Universidad Icesi, Cali, Colombia
Lina Quintero - Department of Information Technology, Faculty of Engineering, Universidad Icesi, Cali, Colombia

Introduction: Cutaneous Leishmaniasis is a neglected tropical disease caused by a parasite. The most common presumptive diagnostic tool for this disease is the visual examination of the associated skin lesions by medical experts. Here, a mobile application was developed to aid this pre-diagnosis using an automatic image recognition software based on a convolutional neural network model.Material and Methods: A total of ۲۰۲۲ images of cutaneous diseases taken from ۲۰۱۲ to ۲۰۱۸ were used for training. Then, in ۲۰۱۹, machine learning techniques were tested to develop an automatic classification model. Also, a mobile application was developed and tested against specialized human experts to compare its performance.Results: Transfer learning using the VGG۱۹ model resulted in a ۹۳% accuracy of the classification model. Moreover, on average, the automatic model performance on a randomly selected skin image sample revealed a ۹۹% accuracy while, the ensemble prediction of seven human medical expert’s accuracy was ۸۳%.Conclusion: Mobile skin monitoring applications are crucial developments for democratizing health access, especially for neglected tropical diseases. Our results revealed that the image recognition software outperforms human medical experts and can alert possible patients. Future developments of the mobile application will focus on health monitoring of Cutaneous Leishmaniasis patients via community leaders and aiming at the promotion of treatment adherence.

Infection, Cutaneous Leishmaniasis, Diagnosis