D0272 - Artificial Intelligence and Anorectal Manometry: Automatic Detection and Differentiation of Anorectal Motility Patterns - A Proof of Concept Study

Miguel Mascarenhas, MD¹, Ilario Froehner, MD, PhD², Tiago Ribeiro, MD¹, Joao Afonso, MD³, Pedro Sousa, PhD⁴, Maria Vila Pouca, PhD⁴, João Ferreira, PhD⁵, Guilherme Macedo, MD, PhD^1
1Centro Hospitalar de S. João, Porto, Porto, Portugal; ²Hospital Marcelino Champagnat, Curitiba, Parana, Brazil; ³Centro Hospitalar de São João, Porto, Porto, Portugal; ⁴FEUP - Faculdade de Engenharia da Universidade do Porto, Porto, Porto, Portugal; ⁵FEUP: Faculdade de Engenharia da Universidade do Porto, Porto, Porto, Portugal

Introduction: Anorectal manometry (ARM) has gained increasing relevance in the evaluation and diagnosis of defecation disorders and anal incontinence, both prevalent in the general population. Despite its usefulness, ARM accessibility is difficulted by the insufficient availability of this exam. Indeed, the complexity of data analysis and the time required for its completion and analysis are significant drawbacks to its clinical availability.

This study aimed to develop and validate a deep learning, artificial intelligence (AI) model to automatically differentiate motility patterns of fecal incontinence (FI) from obstructed defecation (OD), using raw data from ARM.

Methods: Pressure signals were collected from a total of 2469 ARM studies (including 837 patients with anal incontinence and 1189 with obstructed defecation). Both identification and labeling were performed by 2 expert interpreters in ARM and included, besides the reference group, patients with FI and OD. Before training, all signals were resampled by interpolation or by removal of redundant points. The dataset was then split into train and test sets in a patient-based manner, for training and validation respectively. We normalized the training data to avoid data leakage. We then trained and evaluated a deep learning model comprised of a series of 1D Convolutional Neural Networks (1DCNN) followed by a series of Dense layers.

Results: The trained CNN automatically detected and differentiated FI from OD motility patterns with a sensitivity of 84.1%, a specificity of 80.2%, and a precision of 78.1% in a patient-split analysis. Furthermore, The overall accuracy was 85.7%. Patient-split analyses are an important step toward the real-life implementation of deep learning models, mitigating potential biases of its applicability to a clinical setting.

Discussion: Our group developed a pionner AI algorithm for automatic detection and differentiation of relevant anorectal motility patterns. Subsequent development of the CNN as well as more data are required to further develop the model’s diagnostic performance and to incorporate additional manometric diagnoses according to the London classification of anorectal disorders. Nevertheless, this proof of concept study highlights the feasibility of AI analysis in the interpretation and classification of ARM studies. The further development of these tools may optimize the access to ARM studies, which may have a significant impact on the management of patients with anorectal functional diseases.


Disclosures:


Miguel Mascarenhas, MD¹, Ilario Froehner, MD, PhD², Tiago Ribeiro, MD¹, Joao Afonso, MD³, Pedro Sousa, PhD⁴, Maria Vila Pouca, PhD⁴, João Ferreira, PhD⁵, Guilherme Macedo, MD, PhD¹. D0272 - Artificial Intelligence and Anorectal Manometry: Automatic Detection and Differentiation of Anorectal Motility Patterns - A Proof of Concept Study, ACG 2022 Annual Scientific Meeting Abstracts. Charlotte, NC: American College of Gastroenterology.