Self-supervised learning for determining the presence and site of bowel obstruction on upright abdominal radiography
8 affiliations
- 1Department of Gastroenterology, The First Affiliated Hospital of Soochow University, 899 Pinghai Street, Suzhou, Jiangsu 215006 China
- 2Suzhou Clinical Centre of Digestive Diseases, Suzhou, Jiangsu 215006 China
- 3Department of General Surgery, The First Affiliated Hospital of Soochow University, Suzhou, Jiangsu 215006 China
- 4Department of Ultrasound, Institute of Ultrasound in Medicine and Engineering, Zhongshan Hospital, Fudan University, Shanghai, 200032 China
- 5Department of Radiology, The First Affiliated Hospital of Soochow University, Suzhou, Jiangsu 215006 China
- 6Department of Medicine, School of Clinical Medicine, The University of Hong Kong, Pok Fu Lam, Hong Kong SAR China
- 7Department of Geriatric Medicine, Suzhou Kowloon Hospital, Shanghai Jiao Tong University School of Medicine, Suzhou, Jiangsu 215028 China
- 8Department of General Surgery, Jintan Hospital Affiliated to Jiangsu University, Changzhou, Jiangsu 213299 China
Abstract
Background: Detecting and localizing bowel obstruction on upright abdominal radiograph is a challenging task, and ground truth typically relies on imaging examinations, endoscopy or surgical notes within a short time window. The lack of ground truth leads to poor performance on supervised learning. To address this issue, self-supervised learning (SSL) is used to determine the presence and site of bowel obstruction using few labeled samples.
Methods: A total of 1500 labeled and 1922 unlabeled radiographs were categorized based on the ground truth. The SSL model was firstly trained on the unlabeled radiographs, and then fine-tuned on the labeled radiographs. ResNet50 and VGG16 were used for the embedded base encoders, whose weights and parameters were adjusted during training process. Furthermore, it was tested on an external dataset, compared with supervised learning models and human interpreters. Finally, the t-SNE and Grad-CAM were used to visualize the model’s interpretation.
Results: The proposed SSL models showed superior classification performance (VGG16-backboned: accuracy of 0.846, MCC of 0.873, Kappa of 0.770; ResNet50-backboned: 0.833, 0.860, 0.750), outperformed supervised models (Mean: 0.750, 0.792, 0.625) and a senior radiologist (0.748, 0.791, 0.620). Significant statistical differences were observed between VGG16-backboned SSL model and supervised models, as well as human interpreters (P < 0.05). Additionally, the SSL model exhibited better cluster performance through t-SNE and demonstrated outstanding lesion recognition ability through Grad-CAM.
Conclusions: The proposed SSL model exhibited outstanding performance in determining the presence and site of bowel obstruction on upright abdominal radiography based on unlabeled images and limited ground truth, indicating the potential of SSL for computer-aided diagnosis of diseases that are difficult and time-consuming for labeling.
Supplementary Information: The online version contains supplementary material available at https://doi.org/10.1186/s12880-026-02671-y.
Related
- A novel morphological assessment approach for abdominal aortic aneurysms: automated segmentation and measurement using a semi-supervised deep learning algorithm
- Giant Abdominal Mass Caused by Disseminated Hepatic Hydatid Disease Leading to Acute Intestinal Obstruction
- Abdominal cocoon syndrome presenting with high-grade small bowel obstruction and postoperative intra-abdominal collection: a case report from Saudi Arabia
- Aortoduodenal Syndrome: A Ruptured Abdominal Aortic Aneurysm Causing a Small Bowel Obstruction With Stroke-Like Symptoms.
- JEPA-Mamba: a self-supervised joint-embedding architecture with bidirectional Mamba for ECG representation learning
BibTeX
@article{zhu2026selfsupervised,
title = {{Self-supervised learning for determining the presence and site of bowel obstruction on upright abdominal radiography}},
author = {Shiqi Zhu and Ruijie Li and Mingrui Zhu and Zhou Huang and Jiaxi Lin and Jielu Zhou and Yu Wang and Chunfang Xu and Jinzhou Zhu},
journal = {BMC Medical Imaging},
year = {2026},
month = aug,
doi = {10.1186/s12880-026-02671-y},
url = {https://doi.org/10.1186/s12880-026-02671-y},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Self-supervised learning for determining the presence and site of bowel obstruction on upright abdominal radiography
AU - Shiqi Zhu
AU - Ruijie Li
AU - Mingrui Zhu
AU - Zhou Huang
AU - Jiaxi Lin
AU - Jielu Zhou
AU - Yu Wang
AU - Chunfang Xu
AU - Jinzhou Zhu
PY - 2026
DA - 2026/08/10
JO - BMC Medical Imaging
DO - 10.1186/s12880-026-02671-y
UR - https://doi.org/10.1186/s12880-026-02671-y
AB - Background: Detecting and localizing bowel obstruction on upright abdominal radiograph is a challenging task, and ground truth typically relies on imaging examinations, endoscopy or surgical notes within a short time window. The lack of ground truth leads to poor performance on supervised learning. To address this issue, self-supervised learning (SSL) is used to determine the presence and site of bowel obstruction using few labeled samples. Methods: A total of 1500 labeled and 1922 unlabeled radiographs were categorized based on the ground truth. The SSL model was firstly trained on the unlabeled radiographs, and then fine-tuned on the labeled radiographs. ResNet50 and VGG16 were used for the embedded base encoders, whose weights and parameters were adjusted during training process. Furthermore, it was tested on an external dataset, compared with supervised learning models and human interpreters. Finally, the t-SNE and Grad-CAM were used to visualize the model’s interpretation. Results: The proposed SSL models showed superior classification performance (VGG16-backboned: accuracy of 0.846, MCC of 0.873, Kappa of 0.770; ResNet50-backboned: 0.833, 0.860, 0.750), outperformed supervised models (Mean: 0.750, 0.792, 0.625) and a senior radiologist (0.748, 0.791, 0.620). Significant statistical differences were observed between VGG16-backboned SSL model and supervised models, as well as human interpreters (P < 0.05). Additionally, the SSL model exhibited better cluster performance through t-SNE and demonstrated outstanding lesion recognition ability through Grad-CAM. Conclusions: The proposed SSL model exhibited outstanding performance in determining the presence and site of bowel obstruction on upright abdominal radiography based on unlabeled images and limited ground truth, indicating the potential of SSL for computer-aided diagnosis of diseases that are difficult and time-consuming for labeling. Supplementary Information: The online version contains supplementary material available at https://doi.org/10.1186/s12880-026-02671-y.
SN - 1471-2342
ER -