INFORMACIONES PSIQUIÁTRICAS 249

Informaciones Psiquiátricas 2022 - n.º 249 127 Abstract The link between fingerprint generation and central nervous sys- tem growth points to a potential use of fingerprints as risk markers in schizophrenia. However, the high complexity of fingerprint geo- metric patterns requires flexible algorithms capable of characteri- zing such complexity. From an initial sample of fingerprints scan- ned from 612 patients diagnosed with non-affective psychosis and 844 healthy subjects, we have built deep learning classification al- gorithms based on convolutional neural networks. Previously, the general network architecture was chosen from exploratory fittings performed on an independent fingerprint dataset from the Natio- nal Institute of Standards and Technology. The network architecture was then applied for buinding classification algorithms (patients vs. controls) based on single-finger models and multi-finger mo- dels. Classification accuracy estimates were obtained by applying a 5-fold cross-validation scheme. The highest level of accuracy of the single-finger-based networks was achieved by the right thumb network (accuracy = 68%), whereas the highest accuracy of the multi-input models was obtained by the model that simultaneously used images of the left thumb, index and middle fingers (accuracy = 70%). Although the fitted models were based on data from patients with a well-established diagnosis, given that fingerprints remain stable throughout life after birth, our results imply that fingerprints can be applied as early predictors of psychosis. Especially, if used in subpopulations with high prevalence of schizophrenia, such as those at high risk for psychosis. Keywords: schizophrenia, dermatoglyphics, artificial intelligence, deep learning, risk prediction. HUELLAS DACTILARES COMO PREDICTORES DE LA ESQUIZOFRENIA: UN ESTUDIO DE APRENDIZAJE PROFUNDO