Design and validation of an artificial intelligence system to detect the quality of colon cleansing before colonoscopy / Diseño y validación de un sistema de inteligencia artificial para detectar la calidad de la limpieza del colon previa a la realización de la colonoscopia

Background and aims Patients’ perception of their bowel cleansing quality may guide rescue cleansing strategies before colonoscopy. The main aim of this study was to train and validate a convolutional neural network (CNN) for classifying rectal effluent during bowel preparation intake as “adequate” or “inadequate” cleansing before colonoscopy.Patients and methodsPatients referred for outpatient colonoscopy were asked to provide images of their rectal effluent during the bowel preparation process. The images were categorized as adequate or inadequate cleansing based on a predefined 4-picture quality scale. A total of 1203 images were collected from 660 patients. The initial dataset (799 images), was split into a training set (80%) and a validation set (20%). The second dataset (404 images) was used to develop a second test of the CNN accuracy. Afterward, CNN prediction was prospectively compared with the Boston Bowel Preparation Scale (BBPS) in 200 additional patients who provided a picture of their last rectal effluent.ResultsOn the initial dataset, a global accuracy of 97.49%, a sensitivity of 98.17% and a specificity of 96.66% were obtained using the CNN model. On the second dataset, an accuracy of 95%, a sensitivity of 99.60% and a specificity of 87.41% were obtained. The results from the CNN model were significantly associated with those from the BBPS (P<0.001), and 77.78% of the patients with poor bowel preparation were correctly classified.ConclusionThe designed CNN is capable of classifying “adequate cleansing” and “inadequate cleansing” images with high accuracy. (AU)

Antecedentes y objetivos La percepción de los pacientes sobre la calidad de su limpieza intestinal puede guiar las estrategias de limpieza de rescate antes de una colonoscopia. El objetivo principal de este estudio fue entrenar y validar una red neuronal convolucional (CNN) para clasificar el efluente rectal durante la preparación intestinal como «adecuado» o «inadecuado».Pacientes y métodosPacientes no seleccionados proporcionaron imágenes del efluente rectal durante el proceso de preparación intestinal. Las imágenes fueron categorizadas como una limpieza adecuada o inadecuada según una escala de calidad de 4 imágenes predefinida. Se recopilaron un total de 1.203 imágenes de 660 pacientes. El conjunto de datos inicial (799 imágenes) se dividió en un conjunto de entrenamiento (80%) y un conjunto de validación (20%). Un segundo conjunto de datos (404 imágenes) se utilizó para evaluar la precisión de la CNN. Posteriormente, la predicción de la CNN se comparó prospectivamente con la escala de preparación colónica de Boston (BBPS) en 200 pacientes que proporcionaron una imagen de su último efluente rectal.ResultadosEn el conjunto de datos inicial, la precisión global fue del 97,49%, la sensibilidad del 98,17% y la especificidad del 96,66%. En el segundo conjunto de datos, se obtuvo una precisión del 95%, una sensibilidad del 99,60% y una especificidad del 87,41%. Los resultados del modelo de CNN se asociaron significativamente con la escala de preparación colónica de Boston (p<0,001), y el 77,78% de los pacientes con una preparación intestinal deficiente fueron clasificados correctamente.ConclusiónLa CNN diseñada es capaz de clasificar imágenes de «limpieza adecuada» y «limpieza inadecuada» con alta precisión. (AU)

Design and validation of an artificial intelligence system to detect the quality of colon cleansing before colonoscopy.

BACKGROUND AND AIMS: Patients' perception of their bowel cleansing quality may guide rescue cleansing strategies before colonoscopy. The main aim of this study was to train and validate a convolutional neural network (CNN) for classifying rectal effluent during bowel preparation intake as "adequate" or "inadequate" cleansing before colonoscopy. PATIENTS AND METHODS: Patients referred for outpatient colonoscopy were asked to provide images of their rectal effluent during the bowel preparation process. The images were categorized as adequate or inadequate cleansing based on a predefined 4-picture quality scale. A total of 1203 images were collected from 660 patients. The initial dataset (799 images), was split into a training set (80%) and a validation set (20%). The second dataset (404 images) was used to develop a second test of the CNN accuracy. Afterward, CNN prediction was prospectively compared with the Boston Bowel Preparation Scale (BBPS) in 200 additional patients who provided a picture of their last rectal effluent. RESULTS: On the initial dataset, a global accuracy of 97.49%, a sensitivity of 98.17% and a specificity of 96.66% were obtained using the CNN model. On the second dataset, an accuracy of 95%, a sensitivity of 99.60% and a specificity of 87.41% were obtained. The results from the CNN model were significantly associated with those from the BBPS (P<0.001), and 77.78% of the patients with poor bowel preparation were correctly classified. CONCLUSION: The designed CNN is capable of classifying "adequate cleansing" and "inadequate cleansing" images with high accuracy.

Ganglion cell layer analysis with deep learning in glaucoma diagnosis. / Análisis de la capa de células ganglionares con deep learning en el diagnóstico de glaucoma.

OBJECTIVE: To determine and compare the diagnostic precision in glaucoma of two deep learning models using infrared images of the optic nerve, eye fundus, and the ganglion cell layer (GCL). METHODS: We have selected a sample of normal and glaucoma patients. Three infrared images were registered with a spectral-domain optical coherence tomography (SD-OCT). The first corresponds to the confocal scan image of the fundus, the second is a cut-out of the first centered on the optic nerve, and the third was the SD-OCT image of the GCL. Our deep learning models are developed on the MatLab platform with the ResNet50 and VGG19 pre-trained neural networks. RESULTS: 498 eyes of 298 patients were collected. Of the 498 eyes, 312 are glaucoma and 186 are normal. In the test, the precision of the models was 96% (ResNet50) and 96% (VGG19) for the GCL images, 90% (ResNet50) and 90% (VGG19) for the optic nerve images and 82% (ResNet50) and 84% (VGG19) for the fundus images. The ROC area in the test was 0.96 (ResNet50) and 0.97 (VGG19) for the GCL images, 0.87 (ResNet50) and 0.88 (VGG19) for the optic nerve images, and 0.79 (ResNet50) and 0.81 (VGG19) for the fundus images. CONCLUSIONS: Both deep learning models, applied to the GCL images, achieve high diagnostic precision, sensitivity and specificity in the diagnosis of glaucoma.