Centro Hospitalar de São João Porto, Porto, Portugal
Tiago Ribeiro, MD1, Miguel Mascarenhas, MD1, Joao Afonso, MD1, Hélder Cardoso, MD1, Patrícia Andrade, MD1, João Ferreira, PhD2, Marco Parente, PhD2, Renato Natal, PhD2, Guilherme Macedo, MD, PhD, FACG1 1Centro Hospitalar de São João, Porto, Porto, Portugal; 2University of Porto, Porto, Porto, Portugal
Introduction: The diagnosis of gastric polypoid lesions is increasing due to the wide availability of esophagogastroduodenoscopy (EGD). EGD is the gold standard for diagnosis and follow up of gastric protruding lesions. Nevertheless, it is an invasive procedure and patients may not tolerate it. Capsule endoscopy has emerged as a minimally invasive, patient friendly alternative to conventional EGD. The investigation of artificial intelligence (AI) algorithms for application to clinical practice has flourished in the last decade. The application of AI tools for automatic analysis of CE images has provided promising results. Convolutional Neural Networks (CNN) are a multi-layer artificial intelligence architecture with high performance levels for image analysis. The application of these automated algorithms for detection of gastric protruding lesions in CE images has not been explored. This pilot study aimed to develop and test a CNN-based algorithm for automatic detection gastric protruding lesions in CE images.
Methods: A convolutional neural network was developed based on a total of 890 CE images (180 images containing gastric protruding lesions and 710 showing normal mucosa). A training dataset comprising 80% of the total pool of images (n = 712) was used for development of the network. The performance of the CNN was evaluated using an independent validation dataset (20% of total image pool, n = 178). The output provided by the network was compared to a consensus classification provided by two gastroenterologists with experience in CE (Figure 1a). We evaluated the performance of the network by calculating its sensitivity, specificity, accuracy, positive predictive and negative predictive values (PPV and NPV, respectively) and area under the receiver operating characteristic curve (AUC).
Results: After optimizing the architecture of the network, our model automatically detected gastric protruding lesions with an accuracy of 93.8%(Figure 1b). Our CNN had a sensitivity, specificity, PPV and NPV of 83.3%, 96.5%, 85.7%, and 95.8%, respectively. The AUC was 0.96 (Figure 1c). The CNN analyzed the validation dataset in 3 seconds, at a rate of approximately 56 frames per second.
Discussion: We developed a pioneer CNN which detected gastric protruding lesions with high accuracy. The development of these systems may boost the diagnostic efficiency of CE for the detection non-small bowel lesions, thus expanding the indications for its use.
Figure: Figure 1: 1a - Output obtained from the application of the convolutional neural network. A blue bar represents a correct prediction. A red bar represents an incorrect prediction 1b - Evolution of the accuracy of the convolutional neural network during training and validation phases, as the training and validation datasets were repeatedly inputted in the neural network. 1c - ROC analyses for detection of different types of lesions. CNN – convolutional neural network; N – normal mucosa; PR - protruding lesions.
Disclosures: Tiago Ribeiro indicated no relevant financial relationships. Miguel Mascarenhas indicated no relevant financial relationships. Joao Afonso indicated no relevant financial relationships. Hélder Cardoso indicated no relevant financial relationships. Patrícia Andrade indicated no relevant financial relationships. João Ferreira indicated no relevant financial relationships. Marco Parente indicated no relevant financial relationships. Renato Natal indicated no relevant financial relationships. Guilherme Macedo indicated no relevant financial relationships.
Tiago Ribeiro, MD1, Miguel Mascarenhas, MD1, Joao Afonso, MD1, Hélder Cardoso, MD1, Patrícia Andrade, MD1, João Ferreira, PhD2, Marco Parente, PhD2, Renato Natal, PhD2, Guilherme Macedo, MD, PhD, FACG1. P3076 - Artificial Intelligence and Capsule Endoscopy: Automatic Detection of Gastric Protruding Lesions Using a Convolutional Neural Network, ACG 2021 Annual Scientific Meeting Abstracts. Las Vegas, Nevada: American College of Gastroenterology.