Artificial intelligence for early gastric cancer boundary recognition in NBI and nF-NBI endoscopic images

Scandinavian Journal of Gastroenterology, Год журнала: 2025, Номер unknown, С. 1 - 11

Опубликована: Июнь 2, 2025

Precise delineation of early gastric cancer (EGC) margins is essential for complete resection during endoscopic submucosal dissection. This study aimed to develop deep learning-based models EGC boundary detection in narrow-band imaging (NBI) and near-focus NBI (NF-NBI) images. A total 1215 1646 NF-NBI images from patients were used train three convolutional neural networks (CNN1-CNN3), generating six learning (Model1-Model6). Segmentation performance was compared among endoscopists varying seniority. On images, Model3 achieved an accuracy 0.9348, 0.7272, 0.7277, 0.9435 junior, intermediate, senior endoscopists, respectively. The corresponding Dice coefficients 0.8310 (95% CI, 0.8120-0.8500), 0.6153 0.5827-0.6480), 0.6528 0.6237-0.6819), 0.8360 0.8169-0.8550), with recall values 0.9773, 0.6845, 0.7596, 0.9784, Model6 showed 0.9483, 0.6885 (junior), 0.7826 (intermediate), 0.9621 (senior endoscopists). 0.8526 0.8410-0.8642), 0.6757 0.6569-0.6944), 0.7161 0.6941-0.7382), 0.8618 0.8512-0.8725), 0.9831, 0.8095, 0.8317, 0.9889, proposed accurately delineated boundaries achieving diagnostic comparable that endoscopists.

Язык: Английский

---

Benchmarking Polyp Segmentation Methods in Narrow-Band Imaging Colonoscopy Images

IEEE Journal of Biomedical and Health Informatics, Год журнала: 2023, Номер 27(7), С. 3360 - 3371

Опубликована: Апрель 26, 2023

In recent years, there has been significant progress in polyp segmentation white-light imaging (WLI) colonoscopy images, particularly with methods based on deep learning (DL). However, little attention paid to the reliability of these narrow-band (NBI) data. NBI improves visibility blood vessels and helps physicians observe complex polyps more easily than WLI, but images often include small/flat appearances, background interference, camouflage properties, making a challenging task. This paper proposes new dataset (PS-NBI2K) consisting 2,000 pixel-wise annotations, presents benchmarking results analyses for 24 recently reported DL-based PS-NBI2K. The show that existing struggle locate smaller sizes stronger extracting both local global features performance. There is also trade-off between effectiveness efficiency, most cannot achieve best areas simultaneously. work highlights potential directions designing release PS-NBI2K aims drive further development this field.

Язык: Английский

---

Concealed Object Segmentation with Hierarchical Coherence Modeling

Lecture notes in computer science, Год журнала: 2024, Номер unknown, С. 16 - 27

Опубликована: Янв. 1, 2024

Язык: Английский

---

Attention-Guided and Noise-Resistant Learning for Robust Medical Image Segmentation

IEEE Transactions on Instrumentation and Measurement, Год журнала: 2024, Номер 73, С. 1 - 13

Опубликована: Янв. 1, 2024

Язык: Английский

---

ASPS: Augmented Segment Anything Model for Polyp Segmentation

Lecture notes in computer science, Год журнала: 2024, Номер unknown, С. 118 - 128

Опубликована: Янв. 1, 2024

Язык: Английский

---