Towards Polyp Counting In Full-Procedure Colonoscopy Videos

TL;DR

The paper tackles automated polyp counting in full-procedure colonoscopy videos by learning tracklet embeddings through self-supervised contrastive learning with two encoders (SFE and MVE) and specialized temporal sampling. It replaces threshold-based re-identification with unsupervised clustering (notably Affinity Propagation) to re-associate polyp tracklets, achieving robust performance under a fixed false-positive rate. On the open REAL-Colon dataset, the approach yields a fragmentation rate of $FR=6.30$ with false-positive rate below $0.05$ across the test set, substantially reducing fragmentation compared to prior methods. The work provides open data splits and released code to enable replication and future research in automated colonoscopy reporting and quality benchmarking.

Abstract

Automated colonoscopy reporting holds great potential for enhancing quality control and improving cost-effectiveness of colonoscopy procedures. A major challenge lies in the automated identification, tracking, and re-association (ReID) of polyps tracklets across full-procedure colonoscopy videos. This is essential for precise polyp counting and enables automated computation of key quality metrics, such as Adenoma Detection Rate (ADR) and Polyps Per Colonoscopy (PPC). However, polyp ReID is challenging due to variations in polyp appearance, frequent disappearance from the field of view, and occlusions. In this work, we leverage the REAL-Colon dataset, the first open-access dataset providing full-procedure videos, to define tasks, data splits and metrics for the problem of automatically count polyps in full-procedure videos, establishing an open-access framework. We re-implement previously proposed SimCLR-based methods for learning representations of polyp tracklets, both single-frame and multi-view, and adapt them to the polyp counting task. We then propose an Affinity Propagation-based clustering method to further improve ReID based on these learned representations, ultimately enhancing polyp counting. Our approach achieves state-of-the-art performance, with a polyp fragmentation rate of 6.30 and a false positive rate (FPR) below 5% on the REAL-Colon dataset. We release code at https://github.com/lparolari/towards-polyp-counting.

Figures (1)

• Figure 1: Proposed approach. Fig. \ref{['fig:method-encoder']}. The encoder takes a polyp tracklet and computes its representation. Fig. \ref{['fig:method-clustering']}. Pairwise similarities are computed from video tracklets and a clustering algorithm groups together trackelets to form polyp entities.