P5120 - Deep Learning for Detection and Differentiation of Subepithelial Lesions: A Multicentric Transatlantic Endoscopic Ultrasound Study

Miguel Mascarenhas Saraiva, MD, PhD¹, António Miguel M P D. Martins Pinto da Costa, MD², Francisco Mendes, MD¹, Belen Agudo, MD³, Jessica Widmer, DO⁴, Uzma D. Siddiqui, MD, FACG⁵, Tiago Ribeiro, MD¹, Miguel Martins, MD⁶, Pedro Cardoso, MD¹, Maria João Almeida, MD⁶, Joana Mota, MD¹, João Afonso, MD¹, G. E. Kim, MD⁷, Daniel De la Iglesia Garcia, MD, PhD³, Ana Pérez González, MD³, Carlos Esteban Fernández-Zarza, MD³, Ana Garcia Garcia de Paredes, MD⁸, Maria Moris, MD⁹, Matheus Ferreira de Carvalho, MD¹⁰, Marcos Eduardo Lera dos Santos, MD¹⁰, João Ferreira, PhD¹¹, Khoon-Sheng Kok, MD¹², Tamas Gonda, MD¹³, Masson Adams, MD¹⁴, Jack Di Palma, MD, MACG¹⁵, André Martins dos Santos, MD¹⁶, Filipe Vilas Boas, PhD⁶, Pedro Moutinho Ribeiro, MD, PhD⁶, Susana Lopes, MD, PhD⁶, Mariano Gonzalez Haba, MD³, Eduardo Hourneaux De Moura, MD, PhD¹⁷, Guilherme Macedo, MD, PhD¹
1Centro Hospitalar Universitário São João, Porto, Porto, Portugal; 2Hospital Universitario Puerta de Hierro-Majadahonda, Madrid, Madrid, Spain; 3Hospital Universitario Puerta de Hierro Majadahonda, Madrid, Madrid, Spain; 4NYU Langone Health, Mineola, NY; 5University of Chicago Medicine, Chicago, IL; 6Centro Hospitalar Universitário de São João, Porto, Porto, Portugal; 7Center for Endoscopic Research and Therapeutics (CERT), University of Chicago, Chicago, IL; 8Hospital Universitario Ramón y Cajal, Madrid, Madrid, Spain; 9Hospital Universitario Marqués de Valdecilla, GI, Santander, Spain,, Santander, Cantabria, Spain; 10Hospital das Clínicas da Faculdade de Medicina da Universidade de Sao Paulo, São Paulo, Sao Paulo, Brazil; 11Faculdade de Engenharia da Universidade do Porto, Porto, Porto, Portugal; 12Royal Liverpool Hospital, Liverpool, UK, Liverpool, England, United Kingdom; 13NYU Langone Health, New York, NY; 14USA Health, University of South Alabama, Alabama, AL; 15USA Health, University of South Alabama, Mobile, AL; 16Faculdade de Medicina da Universidade do Porto, Porto, Porto, Portugal; 17Hospital das Clínicas da Faculdade de Medicina da USP, São Paulo, Sao Paulo, Brazil

Introduction: Subepithelial lesions (SELs) are commonly identified during esophagogastroduodenoscopy. Nevertheless, their diagnostic approach is complex, and patient management is dependent on appropriate anatomopathological diagnosis. In this context, endoscopic ultrasound offers the possibility of both lesion characterization and sampling, being a first line procedure in patients with suspected upper gastrointestinal tract SELs over 10-millimeter diameter. However, EUS is operator dependent and has a suboptimal diagnostic accuracy. This multicentric study aimed to develop a deep learning model for detection of upper gastrointestinal tract SEL, with distinction between leiomyoma and gastrointestinal stromal tumors (GIST).

Methods: A convolutional neural network (yolo11net) was developed based on 57190 images from 65 patients from across two continents and 5 countries (Portugal, United States of America, Spain, Brazil, United Kingdom). The procedures were performed with 5 different EUS devices. A binary CNN identified and distinguished leiomyoma and GIST. The total data set was divided into a training, validation and testing dataset with a 70%/20%/10% ratio. The model was evaluated through its accuracy and area under receiver-operating characteristic curve (AUC-PR).

Results: Pleomorphic SEL were identified with 91.8% accuracy. The model had an AUC-PR of 0.992 for GIST and 0.982 for leiomyoma.

Discussion: This was the first worldwide study capable of detecting and differentiating the commonest SEL in EUS. AI-driven has potential to revolutionize the approach to SEL, contributing for more accurate differential diagnosis while guiding EUS-guided tissue sampling. The inclusion of data from 5 different EUS devices enhances model interoperability, while integrating data from 5 countries mitigates the risk of demographic bias. Larger multicentric studies are needed for technology implementation.


Figure: Study Design Overview


Figure: Left: Example of CNN detection and classification task; Right: Precision-Recall Curve for CNN

Disclosures:
Miguel Mascarenhas Saraiva indicated no relevant financial relationships.
António Miguel Martins Pinto da Costa indicated no relevant financial relationships.
Francisco Mendes indicated no relevant financial relationships.
Belen Agudo indicated no relevant financial relationships.
Jessica Widmer indicated no relevant financial relationships.
Uzma Siddiqui: Boston Scientific – Consultant, Grant/Research Support, Speakers Bureau. ConMed – Consultant, Speakers Bureau. Cook – Consultant, Speakers Bureau. Medtronic – Consultant, Speakers Bureau. Olympus – Consultant, Grant/Research Support, Speakers Bureau.
Tiago Ribeiro indicated no relevant financial relationships.
Miguel Martins indicated no relevant financial relationships.
Pedro Cardoso indicated no relevant financial relationships.
Maria João Almeida indicated no relevant financial relationships.
Joana Mota indicated no relevant financial relationships.
João Afonso indicated no relevant financial relationships.
G. E. Kim indicated no relevant financial relationships.
Daniel De la Iglesia Garcia indicated no relevant financial relationships.
Ana Pérez González indicated no relevant financial relationships.
Carlos Esteban Fernández-Zarza indicated no relevant financial relationships.
Ana Garcia Garcia de Paredes indicated no relevant financial relationships.
Maria Moris indicated no relevant financial relationships.
Matheus Ferreira de Carvalho indicated no relevant financial relationships.
Marcos Eduardo Lera dos Santos indicated no relevant financial relationships.
João Ferreira indicated no relevant financial relationships.
Khoon-Sheng Kok indicated no relevant financial relationships.
Tamas Gonda indicated no relevant financial relationships.
Masson Adams indicated no relevant financial relationships.
Jack Di Palma indicated no relevant financial relationships.
André Martins dos Santos indicated no relevant financial relationships.
Filipe Vilas Boas indicated no relevant financial relationships.
Pedro Moutinho Ribeiro indicated no relevant financial relationships.
Susana Lopes indicated no relevant financial relationships.
Mariano Gonzalez Haba indicated no relevant financial relationships.
Eduardo Hourneaux De Moura indicated no relevant financial relationships.
Guilherme Macedo indicated no relevant financial relationships.

Miguel Mascarenhas Saraiva, MD, PhD¹, António Miguel M P D. Martins Pinto da Costa, MD², Francisco Mendes, MD¹, Belen Agudo, MD³, Jessica Widmer, DO⁴, Uzma D. Siddiqui, MD, FACG⁵, Tiago Ribeiro, MD¹, Miguel Martins, MD⁶, Pedro Cardoso, MD¹, Maria João Almeida, MD⁶, Joana Mota, MD¹, João Afonso, MD¹, G. E. Kim, MD⁷, Daniel De la Iglesia Garcia, MD, PhD³, Ana Pérez González, MD³, Carlos Esteban Fernández-Zarza, MD³, Ana Garcia Garcia de Paredes, MD⁸, Maria Moris, MD⁹, Matheus Ferreira de Carvalho, MD¹⁰, Marcos Eduardo Lera dos Santos, MD¹⁰, João Ferreira, PhD¹¹, Khoon-Sheng Kok, MD¹², Tamas Gonda, MD¹³, Masson Adams, MD¹⁴, Jack Di Palma, MD, MACG¹⁵, André Martins dos Santos, MD¹⁶, Filipe Vilas Boas, PhD⁶, Pedro Moutinho Ribeiro, MD, PhD⁶, Susana Lopes, MD, PhD⁶, Mariano Gonzalez Haba, MD³, Eduardo Hourneaux De Moura, MD, PhD¹⁷, Guilherme Macedo, MD, PhD¹. P5120 - Deep Learning for Detection and Differentiation of Subepithelial Lesions: A Multicentric Transatlantic Endoscopic Ultrasound Study, ACG 2025 Annual Scientific Meeting Abstracts. Phoenix, AZ: American College of Gastroenterology.