EP142 - Deep learning and High Resolution Anoscopy: A Trinary Interoperable Model for Detecting and Differentiating Clinically Relevant Anal Lesions

Saturday, May 9, 2026

3:31 PM - 3:36 PM ET

ePoster Presenter(s)

PD

Pedro Diaz Donoso, MD (he/him/his)

Colorectal surgeon
Centro Privado de Cirugia y Coloprctologia
Buenos Aires, Ciudad Autonoma de Buenos Aires, Argentina

Disclosure information not submitted.

View ePoster

Author(s)

MM

Miguel Mascarenhas, n/a

Gastroenterology, Centro Hospitalar Universitario de Sao Joao
Hospital São João, Porto.
Porto, Porto, Portugal

All disclosed relationships have been reviewed and managed prior to the start of this activity. Any relevant relationships will be listed below. This individual has no financial relationships to disclose
MM

Miguel Martins, n/a

Engineer
Faculty of Engineering of the University of Porto
Porto, Porto, Portugal

Disclosure information not submitted.
LB

Luis Barroso, .

Wake Forest University Health Sciences, Winston-Salem
.
Wiston-Salem, North Carolina, United States

Disclosure information not submitted.
TM

Thiago Manzione, .

Department of Surgery, Instituto de Infectologia Emílio Ribas
.
., Sao Paulo, Brazil

Disclosure information not submitted.
LL

Luciana La Rosa, .

Centro Privado de Cirugía y Coloproctologia
.
., Ciudad Autonoma de Buenos Aires, Argentina

Disclosure information not submitted.
DC

Dolores Caffarena, .

Centro Privado de Cirugía y Coloproctologia
.
., Ciudad Autonoma de Buenos Aires, Argentina

Disclosure information not submitted.
AA

Amine Alam, MD

Groupe hospitalier Paris Saint-Joseph

Disclosure information not submitted.
JA

Javed Ashamed, .

Department of Surgery, NHS Royal Liverpool University Hospital
.
Liverpool, England, United Kingdom

Disclosure information not submitted.
JF

Joao Ferreira, .

Department of Gastroenterology, São João University Hospital
.
., Porto, Portugal

Disclosure information not submitted.
TR

Tiago Ribeiro, n/a

Department of Surgery, Instituto de Infectologia Emílio Ribas
.
., Porto, Portugal

Disclosure information not submitted.
PC

Pedro Cardoso, MD

Dr
Centro Hospitalar Universitário de São João, Gastroenterology, Porto, Portugal
Porto, Porto, Portugal

Disclosure information not submitted.
FM

Francisco Mendes, .

Department of Gastroenterology, São João University Hospital
.
., Porto, Portugal

Disclosure information not submitted.
MJ

Maria João Almeida, n/a

Department of Gastroenterology, São João University Hospital
.
., Porto, Portugal

Disclosure information not submitted.
JM

Joana Mota, n/a

Dr
Hospital São João, Porto.
Porto, Porto, Portugal

Disclosure information not submitted.
AC

António Costa, MD

Department of Gastroenterology, Puerta del Hierro University Hospital
.
., Madrid, Spain

Disclosure information not submitted.
NF

Nadia Fathallal, .

Department of Proctology, GH Paris Saint-Joseph
.
Paris, Ile-de-France, France

Disclosure information not submitted.
SN

Sidney Nadal, .

Department of Surgery, Instituto de Infectologia Emílio Ribas
.
Sao Paulo, Sao Paulo, Brazil

Disclosure information not submitted.
JF

João Ferreira, .

Department of Mechanical Engineering, Faculty of Engineering of the University of Porto
.
., Porto, Portugal

Disclosure information not submitted.

Senior Author(s)

GM

Guilherme Macedo, .

Department of Gastroenterology, São João University Hospital
.
., Porto, Portugal

Disclosure information not submitted.

Author(s)

VP

Vincent Parades, .

Department of Proctology, GH Paris Saint-Joseph
.
Paris, Ile-de-France, France

Disclosure information not submitted.

Purpose/Background: High-Resolution Anoscopy (HRA) is the gold standard for anal cancer screening. Early detection of premalignant lesions, such as high-grade squamous intraepithelial lesions (HSIL), is essential for reducing anal cancer mortality. However, differentiating HSIL from other HPV-related lesions—such as low-grade squamous intraepithelial lesions (LSIL) and non-dysplastic mimickers (e.g., inflammation)—remains suboptimal, even among experienced clinicians. This study aimed to develop the first model capable of detecting and differentiating anal canal lesions across these three categories

Methods/Interventions: A multicenter, interoperable deep learning model was developed across five clinical sites using a YOLOv11-based object detection architecture using bounding boxes. The model was trained on 191,951 frames extracted from 107 High-Resolution Anoscopy (HRA) procedures. Each frame was annotated based on corresponding histopathological analysis. The dataset was divided into training, validation, and testing subsets. Lesion detection was defined by accurate placement of bounding boxes. The model’s recall (sensitivity), precision (positive predictive value), and accuracy were calculated as weighted averages across five confidence thresholds (0.45, 0.47, 0.50, 0.52, 0.55).

Results/Outcomes: IOverall, the accuracy of the model was 95.2% (94.7- 95.7). For HSIL, recall and precision were 97.9% (97.6–98.2) and 93.9% (92.8–95.1), respectively. For LSIL, recall and precision were 98.9% (98.8–99.0) and 98.5% (98.3–98.7), respectively. For non-dysplastic lesions, recall and precision were 98.7% (98.5–98.8) and 96.0% (94.9–97.2), respectively.

Conclusion/Discussion: The development of a trinary classification model capable of distinguishing HSIL, LSIL, and non-dysplastic mimickers marks a meaningful step toward the clinical adoption of AI-assisted HRA. By enabling accurate and consistent detection and differentiation of clinically relevant anal lesions, this approach has the potential to improve diagnostic performance and support more targeted, precision-based screening strategies.

Trinary interoperable artificial intelligence model for detecting and differentiating anal lesions

Illustrative examples of lesion detection and classification, supported by explainable Al through bounding box overlays

Precisión - recall curve

Precisión - recall curve