Ring-Feature Fusion Network(RFFNet): a deep learning model integrating CNN and transformer with endoscopic distortion correction for high-accuracy real-time diagnosis of inflammatory bowel disease

LI Hualong; WEI Yanling; RUAN Guangcong; MENG Wei; WU Yi; LI Ying; TANG Jiajie; LIU Jingjing; NIAN Yongjian

doi:10.16016/j.2097-0927.202510069

您当前的位置：

首页 >

文章列表页 >

Ring-Feature Fusion Network(RFFNet): a deep learning model integrating CNN and transformer with endoscopic distortion correction for high-accuracy real-time diagnosis of inflammatory bowel disease

Intelligent Medicine and Prediction Model | 更新时间：2026-03-02

- Ring-Feature Fusion Network(RFFNet): a deep learning model integrating CNN and transformer with endoscopic distortion correction for high-accuracy real-time diagnosis of inflammatory bowel disease
- Journal of Army Medical University Vol. 48, Issue 4, Pages: 479-492(2026)
- 作者机构：
  
  1.陆军军医大学（第三军医大学）生物医学工程与影像医学系数字医学教研室，重庆
  2.陆军军医大学（第三军医大学）大坪医院消化内科，重庆
  3.陆军军医大学（第三军医大学）第二附属医院信息科，重庆
- 作者简介：
- 基金信息：
  
  the Chongqing Technology Innovation and Application Development Special Key Project(CSTB2022TIAD-KPX0161);the Chongqing Natural Science Foundation Innovation and Development Joint Fund(CSTB2024NSCQ-LZX0141);the Chongqing Yingcai·Innovation and Entrepreneurship Leading Talent Project(CQYC20220303576)
- DOI：10.16016/j.2097-0927.202510069
  CLC： R319;R445;R574.62
- Received：23 October 2025，
  
  Revised：2025-12-09，
  
  Accepted：16 January 2026，
  
  Published：28 February 2026
- 稿件说明：
移动端阅览
李华龙，魏艳玲，阮广聪，等.环形特征融合网络（RFFNet）：一种融合CNN与Transformer并校正内镜畸变实现炎症性肠病高精度实时诊断的深度学习模型［J］.陆军军医大学学报，2026，48（04）：479-492.DOI：10.16016/j.2097-0927.202510069.

LI H L，WEI Y L，RUAN G C，et al.Ring-Feature Fusion Network（RFFNet）： a deep learning model integrating CNN and transformer with endoscopic distortion correction for high-accuracy real-time diagnosis of inflammatory bowel disease［J］.J Army Med Univ，2026，48（04）：479-492.DOI：10.16016/j.2097-0927.202510069.
李华龙，魏艳玲，阮广聪，等.环形特征融合网络（RFFNet）：一种融合CNN与Transformer并校正内镜畸变实现炎症性肠病高精度实时诊断的深度学习模型［J］.陆军军医大学学报，2026，48（04）：479-492.DOI：10.16016/j.2097-0927.202510069. DOI：

LI H L，WEI Y L，RUAN G C，et al.Ring-Feature Fusion Network（RFFNet）： a deep learning model integrating CNN and transformer with endoscopic distortion correction for high-accuracy real-time diagnosis of inflammatory bowel disease［J］.J Army Med Univ，2026，48（04）：479-492.DOI：10.16016/j.2097-0927.202510069. DOI：

摘要

目的

借助深度学习技术辅助内镜医师对克罗恩病（Crohn’s disease，CD）、溃疡性结肠炎（ulcerative colitis，UC）与正常（Normal）图像肠镜进行准确诊断。

方法

收集2018年1月至2020年11月陆军军医大学大坪医院消化内科与浙江大学医学院附属邵逸夫医院消化内科1 309例受试者共24 492幅肠镜图像，其中包括4 729张CD图像（424例）、7 074张UC图像（605例）与12 689张正常（Normal）图像（280例）。每类按照7∶1∶2的比例将病例随机划分为训练集915例17 136张图像，内部测试集131例2 626张图像，外部验证集263例4 730张图像，分别用于对模型进行训练、测试及验证。针对卷积神经网络（Convolutional Neural Network，CNN）与Transformer网络特征融合问题及肠镜图像畸变问题，本研究基于ResNeSt50与MViTv2构建环形特征融合网络（Ring-Feature Fusion Network，RFFNet）。该网络通过双向通道-空间注意力实现跨阶段特征融合，有效结合CNN局部特征提取与Transformer全局建模优势；引入环形特征融合机制，减轻肠镜图像桶形畸变与景深现象，提升模型对远端肠腔几何特性的适应能力。与现有5种深度学习模型进行对比，通过混淆矩阵展示各模型在外部验证集上的识别结果，利用准确率、灵敏度、特异度、F

值与曲线下面积（area under the curve，AUC）等各项指标验证RFFNet的性能优势；通过消融实验，在外部验证集上利用准确率与F

值等指标验证RFFNet中各改进点的有效性与必要性；通过梯度加权类激活映射（gradient-weighted class activation mapping，Grad-CAM）算法生成类激活映射图，直观地展示模型关注区域。

结果

RFFNet在外部验证集上总体准确率为95.68%（95%

CI：

94.14%~98.58%），总体AUC为0.987（95%

： 0.985~0.990）。CD、UC和正常三类别的AUC分别为0.982（95%

： 0.979~0.986）和0.981（95%

： 0.977~0.984）和0.997（95%

： 0.995~0.998）；灵敏度分别为93.35%（95%

： 91.67%~94.71%）、94.00%（95%

： 92.60%~95.16%）和98.56%（95%

： 97.99%~98.97%）；特异度分别为94.58%（95%

： 93.81%~95.27%）、94.20%（95%

： 93.36%~94.94%）和98.6

1%（95%

： 98.05%~99.01%）；F

值分别为87.81%（95%

： 86.28%~89.25%）、90.05%（95%

： 89.00%~91.25%）和98.58%（95%

： 98.22%~98.92%），表明RFFNet实现了较高精度诊断，与单独使用 CNN、Transformer 模型相比，RFFNet的总体准确率提升显著，差异具有统计学意义（

0.05）。消融实验证实，模型能够通过动态空间注意力机制，深度融合CNN的局部细粒度特征提取能力与Transformer的全局上下文建模优势，与只使用CNN模型相比，CNN-Transformer架构的总体准确率提升了0.11%；能够通过环形特征校正肠镜光学畸变与景深衰减，增强对远端肠腔几何结构的建模能力，使用环形特征后RFFNet总体准确率提升了0.57%。类激活映射图表明改进的模型具有自适应捕获肠镜结构特征的能力。

结论

RFFNet通过双骨干架构深度耦合CNN局部纹理感知与Transformer全局依赖建模，并以环形特征显式校正肠镜畸变，实现了CD、UC与正常黏膜的高精度实时分型，实现了CD、UC和正常图像的高精度实时诊断。

Abstract

Objective

To leverage deep learning technology to assist endoscopists in accurately diagnosing colonoscopy images of Crohn’s disease (CD)

ulcerative colitis (UC)

and Normal tissue.

Methods

A total of 24 492 colonoscopy images from 1 309 subjects were collected from Department of Gastroenterology

Daping Hospital

Army Medical University

and Sir Run Run Shaw Hospital between January 2018 and November 2020. The dataset included 4 729 CD images from 424 cases

7 074 UC images from 605 cases

and 12 689 Normal images from 280 cases. The images were randomly split by case in a 7∶1∶2 ratio into a training set (915 cases

17 136 images)

an internal test set (131 cases

2 626 images)

and an external validation set (263 cases

4 730 images). To address feature fusion between CNN and Transformer architectures and correct endoscopic image distortion

we constructed a Ring-Feature Fusion Network (RFFNet) based on ResNeSt50 and MViTv2. The network employs bidirectional channel-spatial attention for cross-stage feature fusion

effectively combining the local feature extraction strengths of CNN with the global modeling capabilities of Transformer. An innovative ring-feature mechanism was introduced to handle barrel distortion and depth-of-field phenomena in colono

scopy images

enhancing the model’s adaptability to the geometric properties of the distal intestinal lumen. RFFNet was compared with five existing deep learning models. Recognition results on the external validation set were visualized using confusion matrices

and performance was evaluated using accuracy

sensitivity

specificity

-score

and area under the curve (AUC). Ablation studies were conducted on the external validation set using metrics like accuracy and F1-score to validate the effectiveness and necessity of each improvement in RFFNet. Gradient-weighted class activation mapping (Grad-CAM) was used to generate heatmaps

visually demonstrating the model’s focus areas and improving interpretability.

Results

On the external validation set

RFFNet achieved an overall accuracy of 95.68% (95%

: 94.14% to 98.58%) and an overall AUC of 0.987 (95%

: 0.985 to 0.990). For CD

and Normal classes

the AUCs were 0.982 (95%

: 0.979 to 0.986)

0.981 (95%

: 0.977 to 0.984)

and 0.997 (95%

: 0.995 to 0.998)

respectively; sensitivities were 93.35% (95%

: 91.67% to 94.71%)

94.00% (95%

: 92.60% to 95.16%)

and 98.56% (95%

: 97.99% to 98.97%)

respectively; specificities were 94.58% (95%

: 93.81% to 95.27%)

94.20% (95%

: 93.36% to 94.94%)

and 98.61% (95%

: 98.05% to 99.01%)

respectively; F1-scores were 87.81% (95%

: 86.28% to 89.25%)

90.05% (95%

: 89.00% to 91.25%)

and 98.58% (95%

: 98.22% to 98.92%)

respectively. These results indicate that RFFNet achieved high diagnostic precision. Compared to using CNN or Transformer models alone

RFFNet’s overall accuracy improvement was statistically significant (

0.05). Ablation studies confirmed that the model

through a dynamic spatial attention mechanism

eeply integrates CNN’s local fine-grained feature extraction with Transformer’s global contextual modeling advantages. The CNN-Transformer architecture improved overall accuracy by 0.11% compared to using CNN alone

respectively. The ring-feature mechanism corrected endoscopic optical distortion and depth-of-field attenuation

enhancing modeling of the distal lumen geometry

and improved RFFNet’s overall accuracy by 0.57%. Grad-CAM heatmaps demonstrated the model’s adaptive ability to capture structural features in colonoscopy images.

Conclusion

RFFNet

through its dual-backbone architecture that deeply couples CNN’s local texture perception with Transformer’s global dependency modeling

and explicit correction of endoscopic distortion

via

ring-features

enables high-accuracy real-time classification of CD

and normal mucosa.

关键词

Keywords

references

薛文俊，杨嫚，蒋秋焕 . 炎症性肠病患者症状群管理的研究进展［J］. 中国护理管理， 2023 ， 23 （ 2 ）： 298 - 303 . DOI： 10.3969/j.issn.1672-1756.2023.02.028 http://dx.doi.org/10.3969/j.issn.1672-1756.2023.02.028 .

XUE W J ， YANG M ， JIANG Q H . Research progress on symptom cluster management in patients with Inflammatory Bowel Disease ［J］. Chin Nurs Manag ， 2023 ， 23 （ 2 ）： 298 - 303 . DOI： 10.3969/j.issn.1672-1756.2023.02.028 http://dx.doi.org/10.3969/j.issn.1672-1756.2023.02.028 .

ZHANG Y M ， CHU X T ， WANG L ， et al . Global patterns in the epidemiology， cancer risk， and surgical implications of inflammatory bowel disease ［J］. Gastroenterol Rep ， 2024 ， 12 ： goae053 . DOI： 10.1093/gastro/goae053 http://dx.doi.org/10.1093/gastro/goae053 .

LI L L ， CHENG R ， WU Y S ， et al . Diagnosis and management of inflammatory bowel disease ［J］. J Evid Based Med ， 2024 ， 17 （ 2 ）： 409 - 433 . DOI： 10.1111/jebm.12626 http://dx.doi.org/10.1111/jebm.12626 .

GORDON H ， BIANCONE L ， FIORINO G ， et al . ECCO guidelines on inflammatory bowel disease and malignancies ［J］. J Crohns Colitis ， 2023 ， 17 （ 6 ）： 827 - 854 . DOI： 10.1093/ecco-jcc/jjac187 http://dx.doi.org/10.1093/ecco-jcc/jjac187 .

HRACS L ， WINDSOR J W ， GOROSPE J ， et al . Global evolution of inflammatory bowel disease across epidemiologic stages ［J］. Nature ， 2025 ， 642 （ 8067 ）： 458 - 466 . DOI： 10.1038/s41586-025-08940-0 http://dx.doi.org/10.1038/s41586-025-08940-0 .

HUANG T ， ZHAN S Q ， CHEN P J ， et al . Accurate diagnosis of endoscopic mucosal healing in ulcerative colitis using deep learning and machine learning ［J］. J Chin Med Assoc ， 2021 ， 84 （ 7 ）： 678 - 681 . DOI： 10.1097/JCMA.0000000000000559 http://dx.doi.org/10.1097/JCMA.0000000000000559 .

LUO X D ， ZHANG J H ， LI Z G ， et al . Diagnosis of ulcerative colitis from endoscopic images based on deep learning ［J］. Biomed Signal Process Control ， 2022 ， 73 ： 103443 . DOI： 10.1016/j.bspc.2021.103443 http://dx.doi.org/10.1016/j.bspc.2021.103443 .

AGRAWAL M ， ALLIN K H ， PETRALIA F ， et al . Multiomics to elucidate inflammatory bowel disease risk factors and pathways ［J］. Nat Rev Gastroenterol Hepatol ， 2022 ， 19 （ 6 ）： 399 - 409 . DOI： 10.1038/s41575-022-00593-y http://dx.doi.org/10.1038/s41575-022-00593-y .

KIM K O . Endoscopic activity in inflammatory bowel disease： clinical significance and application in practice ［J］. Clin Endosc ， 2022 ， 55 （ 4 ）： 480 - 488 . DOI： 10.5946/ce.2022.108 http://dx.doi.org/10.5946/ce.2022.108 .

BUDZINSKAYA A A ， BELOUSOVA E A ， ORLOVA L P ， et al . Differential diagnosis of inflammatory bowel diseases by endoscopic ultrasound ［J］. Alm Clin Med ， 2023 ， 50 （ 6 ）： 400 - 407 . DOI： 10.18786/2072-0505-2022-50-055 http://dx.doi.org/10.18786/2072-0505-2022-50-055 .

FERNANDEZ C ， GAJIC Z ， ESEN E ， et al . Preoperative risk factors for adverse events in adults undergoing bowel resection for inflammatory bowel disease： 15-year assessment of the American college of surgeons national surgical quality improvement program ［J］. Am J Gastroenterol ， 2023 ， 118 （ 12 ）： 2230 - 2241 . DOI： 10.14309/ajg.0000000000002395 http://dx.doi.org/10.14309/ajg.0000000000002395 .

TONG Y R ， LU K M ， YANG Y Y ， et al . Can natural language processing help differentiate inflammatory intestinal diseases in China？ Models applying random forest and convolutional neural network approaches ［J］. BMC Med Inform Decis Mak ， 2020 ， 20 （ 1 ）： 248 . DOI： 10.1186/s12911-020-01277-w http://dx.doi.org/10.1186/s12911-020-01277-w .

戚婧，阮广聪，杨毅，等 . 空间双线性注意力网络识别溃疡性结肠炎与克罗恩病［J］. 陆军军医大学学报， 2023 ， 45 （ 3 ）： 227 - 234， 242 . DOI： 10.16016/j.2097-0927.202209187 http://dx.doi.org/10.16016/j.2097-0927.202209187 .

QI J ， RUAN G C ， YANG Y ， et al . Identification of ulcerative colitis and Crohn’s disease based on spatial and bilinear attention network ［J］. J Army Med Univ ， 2023 ， 45 （ 3 ）： 227 - 234， 242 . DOI： 10.16016/j.2097-0927.202209187 http://dx.doi.org/10.16016/j.2097-0927.202209187 .

UDRISTOIU A L ， STEFANESCU D ， GRUIONU G ， et al . Deep learning algorithm for the confirmation of mucosal healing in Crohn’s disease， based on confocal laser endomicroscopy images ［J］. J Gastrointestin Liver Dis ， 2021 ， 30 （ 1 ）： 59 - 65 . DOI： 10.15403/jgld-3212 http://dx.doi.org/10.15403/jgld-3212 .

MAURÍCIO J ， DOMINGUES I . Distinguishing between Crohn’s disease and ulcerative colitis using deep learning models with interpretability ［J］. Pattern Anal Appl ， 2024 ， 27 （ 1 ）： 1 . DOI： 10.1007/s10044-023-01206-3 http://dx.doi.org/10.1007/s10044-023-01206-3 .

孙慧贤，罗飞路，张玉华 . 内窥镜图像非线性畸变数字校正方法［J］. 无损检测， 2009 ， 31 （ 2 ）： 92 - 95， 149 .

SUN H X ， LUO F L ， ZHANG Y H . Nonlinear distortion correction of endoscope images with digital technology ［J］. Nondestruct Test Technol ， 2009 ， 31 （ 2 ）： 92 - 95， 149 .

ZHANG H ， WU C R ， ZHANG Z Y ， et al . ResNeSt： split-attention networks ［C］// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops （CVPRW） . June 19-20， 2022 ， New Orleans， LA， USA . IEEE ， 2022 ： 2735 - 2745 . DOI： 10.1109/CVPRW56347. 2022.00309 http://dx.doi.org/10.1109/CVPRW56347.2022.00309 .

LI Y H ， WU C Y ， FAN H Q ， et al . MViTv2： improved multiscale vision transformers for classification and detection ［C］// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR） . New Orleans ： IEEE ， 2022 ： 4794 - 4804 . DOI： 10.1109/CVPR52688.2022.00476 http://dx.doi.org/10.1109/CVPR52688.2022.00476 .

HE K M ， ZHANG X Y ， REN S Q ， et al . Deep residual learning for image recognition ［C］// 2016 IEEE Conference on Computer Vision and Pattern Recognition （CVPR） . Las Vegas ： IEEE ， 2016 ： 770 - 778 . DOI： 10.1109/CVPR.2016.90 http://dx.doi.org/10.1109/CVPR.2016.90 .

DOSOVITSKIY A ， BEYER L ， KOLESNIKOV A ， et al . An image is worth 16 x 16 words： transformers for image recognition at scale［EB/OL］. 2020 ： arXiv ： 2010 . 11929 . https：//arxiv.org/abs/2010.11929 https://arxiv.org/abs/2010.11929 .

LIU Z ， LIN Y T ， CAO Y ， et al . Swin transformer： hierarchical vision transformer using shifted windows ［C］// 2021 IEEE/CVF International Conference on Computer Vision （ICCV） . October 10-17， 2021 ， Montreal， QC， Canada . IEEE ， 2022 ： 9992 - 10002 . DOI： 10.1109/ICCV48922.2021.00986 http://dx.doi.org/10.1109/ICCV48922.2021.00986 .

SELVARAJU R R ， COGSWELL M ， DAS A ， et al . Grad-CAM： visual explanations from deep networks via gradient-based localization ［J］. Int J Comput Vis ， 2020 ， 128 （ 2 ）： 336 - 359 . DOI： 10.1007/s11263-019-01228-7 http://dx.doi.org/10.1007/s11263-019-01228-7 .

YAO T ， LI Y H ， PAN Y W ， et al . HIRI-ViT： scaling vision transformer with high resolution inputs ［J］. IEEE Trans Pattern Anal Mach Intell ， 2024 ， 46 （ 9 ）： 6431 - 6442 . DOI： 10.1109/TPAMI.2024.3379457 http://dx.doi.org/10.1109/TPAMI.2024.3379457 .

GRAHAM B ， ENGELCKE M ， VAN DER MAATEN L . 3D semantic segmentation with submanifold sparse convolutional networks ［C］// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Salt Lake City ： IEEE ， 2018 ： 9224 - 9232 . DOI： 10.1109/CVPR.2018.00961 http://dx.doi.org/10.1109/CVPR.2018.00961 .

MA J ， HE Y T ， LI F F ， et al . Segment anything in medical images ［J］. Nat Commun ， 2024 ， 15 ： 654 . DOI： 10.1038/s41467-024-44824-z http://dx.doi.org/10.1038/s41467-024-44824-z .

WANG Z F ， WU Z B ， AGARWAL D ， et al . MedCLIP： contrastive learning from unpaired medical images and text ［C］// Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing . Abu Dhabi ： United Arab Emirates Association for Computational Linguistics ， 2022 ： 3876 - 3887 .

LUO J B ， LIN J H ， ZHANG Z Y ， et al . UniVid： the open-source unified video model ［EB/OL］. 2025 ： arXiv ： 2509 . 24200 . https：//arxiv.org/abs/2509.24200 https://arxiv.org/abs/2509.24200 .

MORAN G W ， GORDON M ， SINOPOLOU V ， et al . British Society of Gastroenterology guidelines on inflammatory bowel disease in adults： 2025 ［J］. Gut ， 2025 ， 74 （ Suppl 2 ）： s1 - s101 . DOI： 10.1136/gutjnl-2024-334395 http://dx.doi.org/10.1136/gutjnl-2024-334395 .

CHIERICI M ， PUICA N ， POZZI M ， et al . Automatically detecting Crohn’s disease and Ulcerative Colitis from endoscopic imaging ［J］. BMC Med Inform Decis Mak ， 2022 ， 22 （ 6 ）： 300 . DOI： 10.1186/s12911-022-02043-w http://dx.doi.org/10.1186/s12911-022-02043-w .

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Identification of ulcerative colitis and Crohn's disease based on spatial and bilinear attention network

A deep learning-based model for significantly enhancing diagnostic accuracy and efficiency in novice sonographers: development and validation of a YOLOv13-powered real-time abdominal ultrasound anomaly detection system

Artificial intelligence reshaping vaccine development: a new paradigm of efficiency revolution and precision design

Integrative model combining deep learning, clinical and radiomic features enhances EGFR mutation prediction in non-small cell lung cancer

Predictive performance of CT images-based 3D ResNet18 model for identifying lung tuberculosis drug resistance

Related Author

Jing QI

Guangcong RUAN

Yi YANG

Yi WU

Qian CAO

Yanling WEI

Yongjian NIAN

LI Wei

Related Institution

Department of Digital Medicine, School of Biomedical Engineering and Imaging Medicine, Army Medical University(Third Military Medical University)

Department of Gastroenterology, Army Medical Center of PLA

Department of Gastroenterology, Sir Run Run Shaw Hospital, Zhejiang University School of Medicine

Department of Ultrasonography， Second Affiliated Hospital， Army Medical University （Third Military Medical University）

National Engineering Research Center of Immunological Products， Faculty of Pharmacy and Laboratory Medicine，Army Medical University（ Third Military Medical University）

⁰