P1964 - Artificial Intelligence and Capsule Endoscopy: Automatic Detection of Small Bowel Lymphangiectasia and Xanthomas Using a Convolutional Neural Network

Tiago Ribeiro, MD¹, Miguel Mascarenhas, MD¹, Joao Afonso, MD¹, João Ferreira, PhD², Hélder Cardoso, MD¹, Patrícia Andrade, MD¹, Marco Parente, PhD², Renato Natal, PhD², Guilherme Macedo, MD, PhD, FACG^1
1Centro Hospitalar de São João, Porto, Porto, Portugal; ²University of Porto, Porto, Porto, Portugal

Introduction: The detection of enteric lymphangiectasia and xanthomas by capsule endoscopy (CE) exams. Although lymphangiectasia may be a manifestation of numerous diseases (e.g., Whipple’s disease), in most cases, both lymphangiectasia and xanthomas are incidental findings without clear pathological significance. However, their accurate identification is important for a diligent differential diagnosis, as well as serving as topographic references for localization of lesions that may hold clinical and prognostic importance.

Convolutional Neural Networks (CNN) are an artificial intelligence (AI) architecture directed towards image analysis. The application of these algorithms for automated analysis of CE images has provided promising results. However, to date, no study has assessed the ability of CNNs to identify the presence of enteric lymphangiectasia and xanthomas in CE images. We aimed to develop and test a CNN-based algorithm for automatic detection small bowel lymphangiectasiaa and xanthomas.

Methods: A CNN was developed based on a total of 11588 CE images: 1224 images containing lymphangiectasia, 811 showing xanthomas and the remaining showing normal mucosa or other findings. For automatic identification of these findings, these images were inserted into a CNN model with transfer of learning. A training dataset comprising 80% of the total pool of images (n=8271) 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=2317). The output provided by the network was compared to a consensus classification provided by two gastroenterologists with experience in CE (Figure 1). The sensitivity, specificity, accuracy, positive predictive and negative predictive values (PPV and NPV) were calculated.

Results: Our model was able to automatically detect and distinguish the presence of lymphangiectasias and xanthomas with an overall accuracy of 95.5%. The sensitivity, specificity, PPV and NPV were 92.7%, 96.1%, 83.4% and 98.4%, respectively. The CNN complete the analysis of the validation dataset in 22 seconds (approximately 105 images per second).

Discussion: We developed a pioneer deep learning algorithm for detection of lymphangiectasia and xanthomas. Our model demonstrated high performance levels in the detection of lymphangiectasia and xanthomas. The development of accurate AI systems may contribute to improve the diagnostic ability of CE.


Figure: Figure 1: Output obtained from the application of the convolutional neural network. The bars represent the probability estimated by the network. The finding with the highest probability was outputted as the predicted classification. A blue bar represents a correct prediction. N: normal small bowel mucosa; P0L – lymphangiectasia; P0X - Xanthoma. P0 refers to the classification of the bleeding potential according to Saurin's classification.


Disclosures:
Tiago Ribeiro indicated no relevant financial relationships.
Miguel Mascarenhas indicated no relevant financial relationships.
Joao Afonso indicated no relevant financial relationships.
João Ferreira indicated no relevant financial relationships.
Hélder Cardoso indicated no relevant financial relationships.
Patrícia Andrade 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, João Ferreira, PhD², Hélder Cardoso, MD¹, Patrícia Andrade, MD¹, Marco Parente, PhD², Renato Natal, PhD², Guilherme Macedo, MD, PhD, FACG¹. P1964 - Artificial Intelligence and Capsule Endoscopy: Automatic Detection of Small Bowel Lymphangiectasia and Xanthomas Using a Convolutional Neural Network, ACG 2021 Annual Scientific Meeting Abstracts. Las Vegas, Nevada: American College of Gastroenterology.