Self-Supervised Laparoscopic Image Denoising for Endometriosis Detection

In: 2026 International Symposium on Networks, Computers and Communications (ISNCC) · 2026 · pp. 1–8 · doi:10.1109/isncc70543.2026.11693867 · W7213996441
OA: hybrid CC0
⚙ AI-generated summary by qwen3.7-flash, 2026-10-10 ⓘ

This paper presents a self-supervised approach to denoise laparoscopic images, aiming to improve the accuracy of endometriosis detection in surgical settings.

One-sentence paraphrase of the abstract; not a substitute for reading it. No clinical advice. How this works

Abstract

Laparoscopic image quality is often degraded by complex noise patterns, which can negatively impact the visualization of anatomical structures and the detection of subtle lesions such as endometriosis. In practice, obtaining paired clean and noisy images for supervised training is not feasible, making self-supervised approaches essential for real clinical applications. In this paper, we proposed a novel self-supervised framework for laparoscopic In this paper, we propose a hybrid two-stage framework for laparoscopic image denoising. The model first undergoes supervised pre-training on a realistic mixed-noise simulator that approximates Poisson-Gaussian noise, specular reflections, surgical smoke and compression artefacts. It is then fine-tuned in a fully self-supervised manner on real unpaired laparoscopic images using a Self-Supervised Denoising via Data Undersampling (SSDU) strategy. The architecture combines a CNN-based U-Net backbone, a Transformer bottleneck for global context modelling, and a Fourier Neural Operator (FNO) module for explicit frequency-domain restoration. Experimental results demonstrate that the proposed method significantly outperforms classical and existing self-supervised approaches. In addition, the model shows strong generalization across multiple unseen laparoscopic datasets, while preserving fine anatomical details and improving overall image quality. These results highlight the potential of the proposed framework for real-world clinical applications, particularly for improving the reliability of endometriosis detection.

My notes (saved in your browser only)

Citation neighborhood (sparse)

Too few in-corpus citations on either side for a chart; here are the lists.

Cites (2)

References (43)

Source provenance

openalex
last seen: 2026-11-17T06:02:21.439193+00:00
unpaywall
last seen: 2026-11-03T06:29:01.780496+00:00
License: CC0 · commercial use OK