Red neuronal artificial para la detección de edema macular diabético en imágenes de tomografía de coherencia óptica : estudio transversal de prueba diagnóstica
Pinilla Gomez, Carlos Mario
Rosenstiehl, Shirley Margarita
Gomez, Flor Edith
Rodríguez Alvira, Francisco José
Purpose: To calculate the accuracy of “OCT-net”, an artificial neural network, to diagnose diabetic macular edema using optical coherence tomography images. Study design: Cross-sectional diagnostic test study. Methods: we collected 100 image volumes from eyes diagnosed with diabetic macular edema and 100 volumes of healthy individuals, each volume contains 5 B-scan images. We split the images randomly in two datasets: 70% of the images were used to train the artificial neural network and 30% for testing its accuracy. Results: sensitivity was found to be 81,82% (CI 64,54-93,02%) and specificity 88,89% (CI 70,84-97,65%) in the image volume analysis. Positive and negative predictive values were 90% (CI 75,37-96,36%) and 80% (CI 65,71-89,30%) respectively. Conclusions: In our study, OCT-net showed a good performance on detecting diabetic macular edema in optic coherence tomography images. More studies with greater sample size are required.