Endoscopic Image Clustering with Temporal Ordering Information Based on Dynamic Programming.

Area of Science:

• Medical imaging
• Machine learning
• Computer vision

Background:

• Collecting large, manually labeled endoscopic image datasets for machine learning is challenging and costly.
• Clustering endoscopic images offers a group-based labeling approach to mitigate dataset construction expenses.

Purpose of the Study:

• To propose an advanced clustering method for endoscopic image sequences that leverages temporal ordering information.
• To enhance the efficiency and reduce the cost of endoscopic dataset creation for machine learning applications.

Main Methods:

• Utilized a deep neural network for feature extraction from endoscopic images.
• Employed dynamic programming to solve the clustering problem incorporating temporal ordering information.
• Applied the proposed method to cluster esophagogastroduodenoscopy (EGD) images.

Main Results:

• The proposed clustering method demonstrated improved performance by effectively utilizing the sequential properties of endoscopic image data.
• Validation through experiments on esophagogastroduodenoscopy images confirmed the efficacy of the temporal ordering approach.

Conclusions:

• The integration of temporal ordering information significantly enhances the performance of clustering for endoscopic image sequences.
• This method offers a viable solution for cost-effective endoscopic dataset construction, facilitating advancements in machine learning for medical imaging.