Machine Vision System Enables the Inspection of FDA-Approved Medical Guide Wire

Machine Vision System Enables the Inspection of FDA-Approved Medical Guide Wire

Machine Vision System Enables the Inspection of  FDA-Approved Medical Guide Wire
Machine Vision System Enables the Inspection of FDA-Approved Medical Guide Wire

The Challenge


A global medical device manufacturer came to us with the need for a system to automate the inspection of a surgical guidewire.

The Solution


Using Neural Vision and a machine vision system, we ran the customer’s surgical guidewire through an in-depth inspection process that improved their quality control while increasing throughput and decreasing test times.

The Story

The customer’s surgical guidewire is a finely wound coil .3 mm in width used to remove clots in arteries and veins. If the wire has any sharp edges or imperfections, it can lodge itself in the patient’s tissue and cause severe damage. Our customer reached out to us for assistance in developing a new machine vision inspection process for their product, including a custom software application including an operator interface (OE) for use and configuration.


Two Cognex 5MP cameras with two Edmund Optics telecentric lenses attached.
Two Cognex 5MP cameras with two Edmund Optics telecentric lenses attached.

Figure 1. Two Cognex 5MP cameras with two Edmund Optics telecentric lenses attached.


The client provided a mechanical system that rotated the guidewire. Our engineering team paired a Programmable Logic Controller (PLC) with the existing mechanical system to send triggers for the vision system (two Cognex 5MP cameras with two Edmund Optics telecentric lenses attached) to capture images. Using LabVIEW, we then created a graphical user interface for receiving the captured images and saving them for post-processing.

After initial image collection begins, Neural Vision begins analysis through the power of deep learning. It does this by identifying similarities and differences in key characteristics of each image at the pixel level and by learning to make subjective observations over a multitude of images. The software can determine the coil’s length and inspect the shape of the coil’s tip to ensure that it was properly cut, formed, and soldered. The user interface shows a “pass/fail” result for three test cases:


- Guidewire segment is the correct length

- Proper soldering

- Precision scoring of the tip


Initial Prototype.
Initial Prototype.

Figure 2. Initial Prototype.


After the creation of the graphical user interface, the second project phase was to expand and refine Neural Vision’s defect inspection. In the project’s first phase, there was a false reject percentage of 13%. This meant that 13% of the time, the operator identified the part as a pass while Neural Vision determined it a failure. An important goal our team recognized was to reduce the false reject percentage to under 10%.

To accomplish this, we categorized the tip defects rather than giving them all a blanket “bad” label during the inspection. In specifying the tip defects, we tailored the image results to a more consistent accuracy. Our engineers were able to create ten different defect categories with their own individual thresholds. This reduced the frequency of false rejects to 6%.

Overcoming the Obstacles


The largest obstacle our team faced was determining how to accurately tag the differences in characteristics between images for Neural Vision’s success. It was important to decide how to best analyze the results of the guidewire’s tip, and this problem could only be solved through collaboration with the client.

Tagging the images the inspection system collects is the most crucial step in Neural Vision’s development process. This is because Neural Vision software is “programmed” by a user tagging images as “good” or “bad” which pixel-by-pixel builds an accurate model. This requires strong attention to detail as one is often looking for small defects on a high-definition image. If an acceptable image is accidentally tagged as containing a defect, the Neural Vision model can become confused as the data it is being fed is not consistent. The operator who is tagging images must be specific and consistent, over hundreds of images. Each tagged image improves the model’s ability to identify product defects.


Graphical user interface.
Graphical user interface.

Figure 3. Graphical user interface.


Delivering the Outcome


In collaboration with our client, we designed and deployed a vision inspection system that consistently identified good vs. defective wires while reducing the percentage of false rejects. We were able to educate the client on the complete extent of their vision inspection system and Neural Vision’s ability to characterize wire defects. Automating our client’s inspection process saved them significant time and money and improved their quality control process to meet medical standards.



Technical Specifications


• 2 x GigE, 5MP, 24fps, CMOS, Sony IMX264 2/3", Monochrome Cameras

• 2 x 65mm WD Compact Telecentric Len

• 1 x 52mm Telecentric Backlight Illuminator

• 1 x Blue, 0.315" LED Spot Light

• Cyth’s Neural Vision Software

• LabVIEW 2017 (64-bit)