Related Experiment Video
Updated: Jun 28, 2025
11:12
Nondestructive Monitoring of Degradable Scaffold-Based Tissue-Engineered Blood Vessel Development Using Optical Coherence Tomography
Published on: October 3, 2018
5.6K
Automatic and real-time tissue sensing for autonomous intestinal anastomosis using hybrid MLP-DC-CNN classifier-based
Yaning Wang1, Shuwen Wei1, Ruizhi Zuo1
1Department of Electrical and Computer Engineering, Johns Hopkins University, 3400 N. Charles St., Baltimore, MD 21218, USA.
Biomedical Optics Express
|April 18, 2024
Summary
This study introduces an automated tissue classification system using optical coherence tomography (OCT) and a hybrid neural network to detect errors during robotic surgery. The system enhances autonomous robotic surgery by reducing manual monitoring for accurate suturing.
Area of Science:
- Robotic Surgery
- Surgical Technology
- Medical Imaging
Background:
- Anastomosis is a critical surgical procedure in gastrointestinal, urologic, and gynecologic fields.
- Current autonomous surgical robots like the Smart Tissue Autonomous Robot (STAR) system improve efficiency but require manual monitoring to prevent suturing errors.
- Eliminating manual oversight is essential for advancing towards fully autonomous robotic surgery.
Purpose of the Study:
- To develop and integrate an automated tissue classification system with a robotic surgical tool to detect and prevent missed or wrong stitches during anastomosis.
- To enhance the autonomy of the STAR system by removing the need for real-time manual monitoring during suturing.
- To classify abdominal tissue types in real-time using optical coherence tomography (OCT) data.
Main Methods:
- Integrated an optical coherence tomography (OCT) fiber sensor with a robotic suture tool.
- Developed a hybrid multilayer perceptron dual-channel convolutional neural network (MLP-DC-CNN) for real-time tissue classification.
- Utilized handcrafted features (optical properties, morphology) for MLP and intensity-based features (attenuation coefficients) for DC-CNN, combined via decision fusion.
Main Results:
- The MLP-DC-CNN model achieved real-time classification of 1D OCT signals with 90.06% accuracy, 88.34% precision, and 87.29% sensitivity on 69,773 test A-lines.
- The system demonstrated a 300 Hz refresh rate and processed 1,024 A-lines in approximately 1.56 seconds.
- The fully automated tissue sensing model outperformed single classifiers (CNN, MLP, SVM), highlighting the benefit of complementary feature sets and architectures.
Conclusions:
- The developed automated tissue sensing model significantly enhances the capability of robotic surgical systems like STAR.
- This technology has the potential to reduce manual intervention in laparoscopic surgery, paving the way for fully autonomous procedures.
- The hybrid MLP-DC-CNN approach effectively classifies intestinal tissue types, improving the safety and efficiency of robotic anastomosis.
