A Trust-Based Methodology to Evaluate Deep Learning Models for Automatic Diagnosis of Ocular Toxoplasmosis from Fundus Images.

In the automatic diagnosis of ocular toxoplasmosis (OT), Deep Learning (DL) has arisen as a powerful and promising approach for diagnosis. However, despite the good performance of the models, decision rules should be interpretable to elicit trust from the medical community. Therefore, the development of an evaluation methodology to assess DL models based on interpretability methods is a challenging task that is necessary to extend the use of AI among clinicians. In this work, we propose a novel methodology to quantify the similarity between the decision rules used by a DL model and an ophthalmologist, based on the assumption that doctors are more likely to trust a prediction that was based on decision rules they can understand. Given an eye fundus image with OT, the proposed methodology compares the segmentation mask of OT lesions labeled by an ophthalmologist with the attribution matrix produced by interpretability methods. Furthermore, an open dataset that includes the eye fundus images and the segmentation masks is shared with the community. The proposal was tested on three different DL architectures. The results suggest that complex models tend to perform worse in terms of likelihood to be trusted while achieving better results in sensitivity and specificity.

Automatic Diagnosis of Ocular Toxoplasmosis from Fundus Images with Residual Neural Networks.

Ocular toxoplasmosis (OT) is commonly diagnosed through the analysis of fundus images of the eye by a specialist. Despite Deep Learning being widely used to process and recognize pathologies in medical images, the diagnosis of ocular toxoplasmosis(OT) has not yet received much attention. A predictive computational model is a valuable time-saving option if used as a support tool for the diagnosis of OT. It could also help diagnose atypical cases, being particularly useful for ophthalmologists who have less experience. In this work, we propose the use of a deep learning model to perform automatic diagnosis of ocular toxoplasmosis from images of the eye fundus. A pretrained residual neural network is fine-tuned on a dataset of samples collected at the medical center of Hospital de Clínicas in Asunción, Paraguay. With sensitivity and specificity rates equal to 94% and 93%,respectively, the results show that the proposed model is highly promising. In order to replicate the results and advance further in this area of research, an open data set of images of the eye fundus labeled by ophthalmologists is made available.

Variabilidad del espesor de la capa de fibras nerviosas peripapilares entre grupos según severidad de miopía y grupo control / Variability of the peripapillary nerve fiber layer thickness between groups according to severity of myopia and control group

Introducción: La miopía es uno de los errores refractivos más prevalentes, a nivel global 28,3%. La apariencia morfológica del disco óptico en la miopía con las variaciones propias de la condición hacen del diagnóstico clínico y el monitoreo de la progresión del glaucoma muy difícil. Recientemente, nuevas modalidades de imagen de alta resolución se desarrollaron para darnos más información y ayudar en la detección de los cambios micro anatómicos de la cabeza del nervio óptico. Objetivo: Determinar la variabilidad del espesor de la capa de fibras nerviosas peripapilares (CFNR) medido con tomografía de coherencia óptica (OCT) entre los diferentes grupos según severidad de miopía y grupo control. Metodología: Se trata de un estudio observacional, descriptivo, de corte transversal. Muestreo no probabilístico de casos consecutivos. Con componente analítico. Resultados: Se analizaron un total de 144 ojos, de los sujetos estudiados aproximadamente dos tercios eran del sexo femenino (n: 97 ojos). La edad media de los sujetos estudiados fue de 30,68 +/- 11,58. Se estudiaron 50 (34,7%) ojos emétropes como grupo control, entre los diferentes grupos según severidad de la miopía se estudiaron 27 (18,8 %) ojos con miopía leve, 33 (22,9 %) con miopía moderada y 34 (23,6 %) ojos con miopía severa. La media de longitud axial encontrada fue de 23,12 +/- 0,81mm para el grupo control (emétropes); 23,72 +/- 0,76 para el grupo con miopía leve; 25,11+/-0.79mm para el grupo con miopía moderada y de 26,42 +/- 1,45mm para el grupo con miopía severa, encontrándose diferencias estadísticamente significativas entre estos grupos con una longitud axial mayor para el grupo con miopía severa. Se encontraron diferencias estadísticamente significativas entre los grupos estudiados, para diferencias encontradas en el espesor promedio y en el espesor de los cuadrantes superior, inferior y temporal. Las diferencias encontradas en el espesor del cuadrante nasal fueron no significativas. Conclusión: A pesar de las limitaciones, este estudio claramente demuestra que los ojos con miopía severa tienen un espesor menor en la CFNR peripapilares que los ojos emétropes en los cuadrantes superior e inferior. En el cuadrante temporal y nasal se encontró un aumento del espesor, siendo estadísticamente significativo en el cuadrante temporal. Por un lado este adelgazamiento podría ser un factor de riesgo para el desarrollo del glaucoma ya que las variaciones en la disposición de las fibras nerviosas de la cabeza del nervio óptico se postulan que hacen del ojo miope más susceptible al daño glaucomatoso. Para la relevancia clínica, debemos tener en cuenta las diferencias propias de la miopía para realizar el diagnóstico correcto en los casos de sospecha de glaucoma y evaluar progresión con OCT. Así también sería ideal agrupar según rango etario por las diferencias propias de esta variable que podría haber influenciado en la interpretación de los resultados.

Introduction: Myopia is one of the most prevalent refractive errors, globally 28.3%. The morphological appearance of the optic disc in myopia with the variations of the condition make clinical diagnosis and monitoring of glaucoma progression very difficult. Recently, new high resolution imaging modalities were developed to give us more information and help in the detection of micro anatomic changes of the optic nerve head. Objective: To determine the thickness variability of the peripapillary nerve fiber (RNFL) layer measured with optical coherence tomography (OCT) between the different groups according to myopia severity and control group. Methodology: This is an observational, descriptive, cross-sectional study. Non-probabilistic sampling of consecutive cases. With analytical component. Results: A total of 144 eyes were analyzed, approximately two thirds of the subjects studied were female (n: 97 eyes). The average age of the subjects studied was 30.68 +/- 11.58. 50 (34.7%) emmetropic eyes were studied as a control group, among the different groups according to severity of myopia 27 (18.8%) eyes with mild myopia were studied, 33 (22.9%) with moderate myopia and 34 (23.6%) eyes with severe myopia. The mean axial length found was 23.12 +/- 0.81mm for the control group (emétropes); 23.72 +/- 0.76mm for the group with mild myopia; 25.11 +/- 0.79mm for the group with moderate myopia and 26.42 +/- 1.45mm for the group with severe myopia, with statistically significant differences between these groups with a greater axial length for the group with severe myopia. Statistically significant differences were found between the groups studied, for differences found in the average thickness and in the thickness of the upper, lower and temporal quadrants. The differences found in the thickness of the nasal quadrant were not significant. Conclusion: Despite the limitations, this study clearly demonstrates that eyes with severe myopia have a smaller thickness in the peripapillary RNFL than the emmetropic eyes in the upper and lower quadrants. An increase in thickness was found in the temporal and nasal quadrant, being statistically significant in the temporal quadrant. On the one hand this thinning could be a risk factor for the development of glaucoma since variations in the arrangement of the nerve fibers of the optic nerve head are postulated that make the myopic eye more susceptible to glaucomatous damage. For clinical relevance, we must take into account the differences of myopia to make the correct diagnosis in cases of suspected glaucoma and evaluate progression with OCT. Thus it would also be ideal to group according to age range due to the differences of this variable that could have influenced the interpretation of the results.