Cutaneous leishmaniasis is a skin disease caused by flagellate protozoa of the genus Leishmania and transmitted by sandflies of the genus Lutzomyia. Around 1 million new cases occur in the world annually, with a total of 12 million people affected, mainly in rural areas with low access to health services and adequate treatments. In the area of the Americas, Colombia has one of the highest infection rates after Brazil. Topical treatments with pentamidine isethionate (PMD) present an attractive alternative due to their ease of application and low costs. However, cutaneous leishmaniasis lesions present nodules with seropurulent exudate that, when drying, form hyperkeratotic lesions, hindering the effective penetration of drugs for their treatment. The use of molecular histology techniques, such as MALDI-MSI, allow in situ evaluation of the penetration of the treatment to the sections of the dermis where the disease-causing parasite resides. However, the large volume of information generated makes it impossible to process it manually. Machine learning techniques allow the unsupervised processing of large amounts of information, generating prediction models for the classification of new information. This work proposes a low-cost method to generate cutaneous leishmaniasis detection and classification models using MALDI-MSI images taken from murine models. The proposed models allow a 95% efficiency when separating healthy samples from infected samples and an effectiveness of 67% when separating effectively treated samples from unsuccessfully treated samples.
Keywords: MALDI-MSI; cutaneous leishmaniasis; machine learning; murine model.