Development of microcatheter tube extrusion angle estimation system using convolutional neural network segmentation

Area of Science:

• Manufacturing Engineering
• Medical Device Technology
• Artificial Intelligence in Manufacturing

Background:

• Microcatheter tubes require stringent quality control due to their medical application.
• Operator variability can lead to inconsistencies in the extrusion angle, impacting microcatheter quality.
• Real-time estimation of extrusion angles during resin extrusion has been a significant challenge.

Purpose of the Study:

• To develop and validate a deep learning-based system for real-time extrusion angle estimation in microcatheter manufacturing.
• To enhance quality control and reduce variability in the microcatheter extrusion process.
• To establish a novel method for real-time extrusion angle measurement.

Main Methods:

• A system employing two RGB cameras for front and side views was designed.
• A convolutional neural network (CNN) was trained for image segmentation of the extruded resin.
• Principal Component Analysis (PCA) was applied to segmented images for accurate extrusion angle estimation.

Main Results:

• The system achieved a mean intersection over union (mIoU) of 0.8848 for segmentation accuracy.
• The mean absolute angle error (MAE) was 0.5968, demonstrating high precision.
• An inference time of 0.0546 seconds confirmed the system's real-time capability.

Conclusions:

• The proposed deep learning system effectively estimates extrusion angles in real-time.
• The validated system is suitable for improving quality control in microcatheter tube manufacturing.
• This represents the first real-time deep learning method for estimating extrusion angles in this process.