Background. The use of mobile applications in dermatology to support remote diagnosis is gaining acceptance, particularly in rural areas, where dermatology services are commonly managed by healthcare personnel with no specialty training. Moreover, ontologies—sets of concepts that represent knowledge in a given domain—are increasingly being used to support medical diagnosis. A specific case is ONTODerm: an ontology to aid dermatological diagnosis. However, there is little information on the combined use of mobile applications and ontologies as support solutions in dermatology. Objective. Assessing the reliability of ONTODerm as a tool to support remote dermatological diagnosis when used together with a mobile dermatological application in underprivileged areas. Methods. A mobile application that allows characterization of skin lesions was developed, and the information about the lesions was sent to ONTODerm. An exploratory study was conducted in a remote area without access to a dermatologist. A total of 64 dermatological queries were recorded in the application and consulted with ONTODerm. Later, an experienced dermatologist evaluated the characterization and diagnosis of each query to determine the accuracy of the system. Results. The results showed that the probability of obtaining a correct diagnosis was between 64.4% and 85.6% with a confidence interval of 95%. A higher accuracy rate was obtained when the skin lesion occurred on the face or when its border was categorized as poorly demarcated. Conclusions. This study demonstrates the implementation of a teledermatology strategy based on mobile applications and domain ontology-driven knowledge base to provide timely assistance to healthcare professionals. This approach was found to be pertinent in the Colombian rural context, particularly in forest regions, where dermatology specialists are not available. The results of this article do not represent a final validation of the proposed approach; they suggest how the ontology can be improved to effectively support medical staff in marginalized regions.
Robustness of convolutional neural networks in recognition of pigmented skin lesions
