2022-08-302022-08-302021-10-08FARIAS, Flávia Monteiro. Uso de aprendizado de máquina para diferenciar dados morfométricos retiniano quanto ao sexo da pessoa. Orientador: Givago da Silva Souza. 2022. 34 f. Trabalho de Curso (Licenciatura em Ciências Biológicas) – Faculdade de Ciências Biológicas, Universidade Federal do Pará, Belém, 2021. Disponível em: https://bdm.ufpa.br:8443/jspui/handle/prefix/4337. Acesso em:.https://bdm.ufpa.br/handle/prefix/4337The present research compared the classification accuracies obtained with the implementation of different K values (2 to 10) of the machine learning algorithm k-nearest neighbors (kNN) in classifying the thickness and volume values of the retinal layers as belonging to male participants or to female participants. The objective is to evaluate whether the K parameter of the kNN algorithm interferes in the classification of sexual dimorphism present in the retinal morphometric data. Data acquisition was performed in the retinal macula with optical coherence tomography in the spectral domain (Spectralis HRA+OCT tomograph) in 64 people (38 women and 26 men) with normal vision, normal or corrected visual acuity ≤ 20/40, without eye diseases or systemic diseases, belonging to the age group of 20 to 40 years. The machine learning analysis considered as characteristics of the data the thickness in the nine regions of the ETDRS and the total macular volume of each retinal layer and as classes the female and male sex. One-way ANOVA and Tukey HSD post-hoc were used for statistical comparisons on the accuracy measures obtained with different implementations of the kNN algorithm with different k parameters, considering a significance level of < 0.05. The kNN algorithm classifies more accurately (> 0.70) the innermost layers of the retina (CFRN, CCG, CNI, CPI, inner retina) and total retina, where we observed significant differences (p < 0.05) between genders, when compared to the classification accuracy (> 0.60) of the layers that do not have significant differences between genders (CPE, CNE, EPR and external retina). The use of different K values in the implementation of the kNN algorithm presents a significant interaction effect F = 2.20 (p = 0.03) between the mean classification accuracy of the total retina. On the other hand, no significant differences were found between the accuracies of the different k values obtained using the morphometric data of the different layers of the retina (p > 0.05). The use of different K values of the kNN algorithm does not affect the performance of the algorithm in classifying retinal layer thickness and volume measurements as belonging to males or females.Acesso AbertoTécnicas de diagnóstico oftalmológicoDimorfismo sexualOftalmopatiasTomografia de coerência opticaDoenças retinianasCNPQ::CIENCIAS BIOLOGICASTrabalho de Curso - Graduação - Monografia