Related Experiment Video
Updated: Aug 10, 2026
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
An Organ-Guided Lightweight Multi-frame Integration Network for Real-Time Abdominal Lesion Detection in Ultrasound
Xinyi Wang1, Mingliu Zhu2, Yaoxian Zou2
1Department of Biomedical Engineering, School of Life Science and Technology, Key Laboratory of Biomedical Information Engineering of the Ministry of Education, Xi'an Jiaotong University, Western China Science & Technology Innovation Harbour, Xi'an, Shaanxi, P.R. China.
A new Organ-Guided Lightweight Multi-frame Integration (OGLMFI) framework improves real-time lesion detection in abdominal ultrasound videos. This deep learning approach enhances accuracy for conditions like gallbladder stones, aiding clinical diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer-Aided Diagnosis
Background:
- Abdominal ultrasound is crucial for screening liver, gallbladder, and kidney diseases.
- Operator dependency and low-contrast imaging challenge real-time lesion detection in ultrasound videos.
- Existing deep learning methods often focus on single organs or static images, limiting real-time multi-organ applicability.
Purpose of the Study:
- To develop a unified, real-time deep learning framework for detecting lesions across multiple abdominal organs in ultrasound videos.
- To address the limitations of operator dependency and improve lesion identification in challenging ultrasound conditions.
Main Methods:
- Proposed the Organ-Guided Lightweight Multi-frame Integration (OGLMFI) framework based on YOLOv11.
- Incorporated an Organ-Guided Feature Filtering module using organ segmentation priors to enhance lesion discrimination.
- Implemented a Lightweight Multi-frame Integration module with a dual-branch fusion strategy for efficient temporal information integration from consecutive frames.
Main Results:
- OGLMFI achieved a Recall of 0.691, mAP50 of 0.703, and mAP50-95 of 0.510 on a test set of 205 clinical videos.
- Demonstrated significant improvements over the YOLOv11-L baseline, with a 10.7% increase in Recall and 8.1% in mAP50-95, while maintaining 52.3 fps real-time inference.
- Achieved the highest performance metrics among evaluated methods and superior average precision across all seven lesion categories, including gallbladder stones and polyps.
Conclusions:
- The OGLMFI framework effectively improves real-time lesion detection in abdominal ultrasound videos through integrated organ-guided filtering and multi-frame fusion.
- Offers a practical, unified solution for detecting lesions across the liver, gallbladder, and kidney.
- Presents a valuable computer-aided tool for enhancing routine abdominal ultrasound screening.
