Keyframe Extraction From Laparoscopic Videos via Diverse and Weighted Dictionary Selection

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Laparoscopic surgery is widespread, generating vast video data for training and quality assurance.
• Manually reviewing these videos is time-consuming, limiting the utility of video archives.
• Efficient summarization is needed to unlock the potential of laparoscopic video data.

Purpose of the Study:

• To develop an automated method for laparoscopic video summarization.
• To extract representative and high-quality keyframes for rapid video access.
• To improve the exploitation of laparoscopic video archives.

Main Methods:

• Utilized deep features from a convolutional neural network for frame representation.
• Formulated summarization as a diverse and weighted dictionary selection model.
• Incorporated image quality and diversity regularization for keyframe selection.
• Developed an iterative algorithm for efficient model optimization.

Main Results:

• The proposed method significantly outperforms existing techniques on a laparoscopic dataset.
• Achieved superior keyframe extraction in terms of quality and diversity.
• Demonstrated rapid convergence of the optimization algorithm.

Conclusions:

• The deep feature-based dictionary selection method offers effective laparoscopic video summarization.
• This approach enhances access to critical surgical information.
• Facilitates surgical training, patient communication, and case file archiving.