Related Experiment Video
Updated: Sep 20, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.0K
Feature discretization-based deep clustering for thyroid ultrasound image feature extraction.
Ruiguo Yu1, Yuan Tian1, Jie Gao1
1College of Intelligence and Computing, Tianjin University, Tianjin, 300350, China; Tianjin Key Laboratory of Cognitive Computing and Application, Tianjin, 300350, China; Tianjin Key Laboratory of Advanced Networking, Tianjin, 300350, China.
Computers in Biology and Medicine
|June 6, 2022
Summary
This study introduces Feature Discretized-based Deep Clustering (FDDC), an unsupervised method for analyzing ultrasound images. FDDC enhances diagnostic accuracy by improving feature representation in deep learning models without requiring image labels.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer-Aided Diagnosis
Background:
- Ultrasound imaging is a popular diagnostic tool due to its convenience and safety.
- Supervised learning for medical image analysis is hindered by high labeling costs.
- Unsupervised methods are trending for Computer-Aided Diagnosis (CAD) but often neglect feature representation.
Purpose of the Study:
- To propose a novel unsupervised deep clustering algorithm, Feature Discretized-based Deep Clustering (FDDC).
- To enhance the representational capability of models in the feature space for ultrasound image analysis.
- To address limitations of existing unsupervised CAD techniques.
Main Methods:
- Introduced Feature Discretized-based Deep Clustering (FDDC) incorporating representation learning.
- Implemented a global-local regular discretization method to constrain feature values and improve expressiveness.
- Utilized a greedy-based label reassignment method to stabilize loss fluctuations during re-clustering.
Main Results:
- FDDC achieved satisfactory results on six classification tasks.
- Demonstrated tumor classification accuracy of 79.06% and machine classification accuracy of 96.17%.
- Outperformed existing unsupervised baseline methods in experimental evaluations.
Conclusions:
- FDDC effectively improves deep clustering for ultrasound image analysis.
- The method enhances model representational capability in feature space.
- FDDC offers a promising unsupervised approach for CAD, overcoming labeling challenges.
