Related Experiment Video
Updated: Sep 10, 2025
10:51
Frame-by-Frame Video Analysis of Idiosyncratic Reach-to-Grasp Movements in Humans
Published on: January 15, 2018
8.5K
Kinematic Adaptive Frame Recognition (KAFR): A Novel Framework for Video Segmentation via Frame Similarity and
Huu Phong Nguyen1, Shekhar Madhav Khairnar1, Sofia Garces Palacios1
1Department of Surgery, University of Texas Southwestern Medical Center, Dallas, TX 75390, USA.
IEEE Access : Practical Innovations, Open Solutions
|August 25, 2025
Summary
This study introduces Kinematics Adaptive Frame Recognition (KAFR), a novel AI technique that significantly reduces surgical video data size and computation time. KAFR improves AI model accuracy by intelligently selecting essential frames for analysis.
Area of Science:
- Artificial Intelligence in Surgery
- Surgical Video Analysis
- Machine Learning for Medical Procedures
Background:
- Surgical videos are crucial for AI-driven analysis and performance assessment.
- The lengthy duration of surgical videos presents a significant challenge for AI model training and efficiency.
- Increasing volumes of surgical video data necessitate advanced techniques for effective processing.
Purpose of the Study:
- To propose and evaluate a novel technique, Kinematics Adaptive Frame Recognition (KAFR), for efficient surgical video frame reduction.
- To reduce dataset size and computation time while preserving and enhancing AI model accuracy.
- To adapt KAFR for surgical phase segmentation and assess its applicability to broader surgical datasets.
Main Methods:
- Developed Kinematics Adaptive Frame Recognition (KAFR) to eliminate redundant frames by tracking surgical tool kinematics.
- Utilized a YOLOv8 model for tool detection, followed by a similarity phase to compute frame variations based on tool movement.
- Trained an X3D Convolutional Neural Network (CNN) for classification, evaluating KAFR on Gastrojejunostomy (GJ) and Pancreaticojejunostomy (PJ) datasets.
Main Results:
- KAFR achieved a tenfold reduction in frames for the GJ dataset, improving accuracy by 4.32% and F1 score by 0.16%.
- On the PJ dataset, KAFR resulted in a fivefold data reduction, with accuracy increasing by 2.05% and F1 score by 2.54%.
- The approach demonstrated competitive performance and efficiency compared to state-of-the-art methods.
Conclusions:
- Kinematics Adaptive Frame Recognition (KAFR) effectively reduces surgical video data size and computation time while improving AI model accuracy.
- KAFR is a versatile technique applicable to various surgical datasets and phase segmentation tasks.
- KAFR can enhance existing AI models by optimizing data input, making it a valuable supplement for surgical data analysis.
