Related Experiment Video
Updated: Aug 6, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
A CNN-based liver ultrasound plane recognition system with multimodal large language model-generated interpretable
Yongjian Chen1, Jingyun Li2, Jiansong Zhang3
1Department of Ultrasound, Second Affiliated Hospital of Fujian Medical University, Quanzhou, China.
Abdominal Radiology (New York)
|July 22, 2026
Summary
A new hybrid AI system combining CNN image recognition and MLLM interpretation effectively identifies liver ultrasound standard planes. This AI tool shows potential for improving image quality control and standardized training in ultrasound diagnostics.
Area of Science:
- Medical Imaging AI
- Ultrasound Diagnostics
- Machine Learning in Healthcare
Background:
- Standardized liver ultrasound planes (LUSPs) are crucial for accurate diagnosis.
- Current training methods for identifying LUSPs can be inconsistent.
- AI offers potential solutions for improving image quality and training efficiency.
Purpose of the Study:
- To develop a hybrid AI pipeline integrating CNNs and MLLMs.
- To assess the AI pipeline's performance in identifying LUSPs.
- To evaluate the AI's utility as an auxiliary tool for quality control and training.
Main Methods:
- A multicenter study utilized 16,782 ultrasound images.
- A hybrid AI model combined CNN for image recognition and MLLM for interpretation.
- Diagnostic performance was compared between radiologists of varying experience levels, students, the AI model, and AI-assisted readers.
Main Results:
- The AI model achieved 88.8% accuracy in recognizing LUSPs and non-standard planes.
- AI performance in identifying liver ultrasound planes was comparable to senior radiologists.
- The hybrid AI system significantly improved diagnostic performance for less experienced readers.
Conclusions:
- The hybrid AI system shows promise as an auxiliary tool for image quality control.
- This AI approach can aid in standardized training for liver ultrasound interpretation.
- Further prospective validation is recommended for this AI system.