(click to enlarge) An automated process makes it easier to spot broken or cracked struts, a bad laser cut, or an uneven electropolish.
Stents are flexible, which presents a challenge for any type of optical inspection. With a printed circuit board or a silicon wafer, for example, each feature is a known absolute, and an electronic test list can be created to easily queue up each detail. Such a list can’t be created with optical equipment for stent inspection. Although some of these inspection tools can automatically manipulate the stent under optics, such tools still require a high level of operator expertise, often for manual fine-tuning of the position or focus. Fully automated routines require especially complicated programming to track the stent as the camera moves along.
Even the best optics typically have a trade-off. Either the image is highly focused and detailed using a close-range lens, or the image is from a wide-view lens that offers more visual area with less-precise detail. This issue has made it difficult to move visual defect detection away from the traditional stereo-microscope. Because stents are cylindrical and the field of the camera is flat, only the central portion of the image is typically in sufficient focus. Also, creating a complete image is very time-consuming and requires a stepping process (also known as quilting).
For small production quantities or aggressive acceptable quality level (AQL) sampling, a partial stent inspection can be adequate for dimensional inspection. However, for the dimensional inspection required for initial development or for stents with highly intricate patterns requiring large amounts of data, these optical tools don’t typically offer the throughput or repeatable accuracy required. Also, the labor cost rises dramatically if manual measurements are involved.
A cardiovascular stent has hundreds of critical features with tight tolerances. Some stents require 100% inspection. Even the best traditional optical CNC inspection requires between 5 and 15 minutes per stent for a complete inspection, not counting setup and programming.
Therefore, most stent inspections still consist of multiple operators viewing individual stents using optical tools and stereomicroscopes. The process is labor-intensive and, from a practical standpoint, open to human error. Even the best operators are prone to fatigue and what is known as saccadic masking—when the brain blocks images as the eye moves rapidly from point to point. Regardless of how well trained and dedicated, a person continually looking into a microscope has saccadic flashes when vision is frozen. Often parts are being rotated under the microscope while the person is looking from point to point. Addressing this problem has been a challenge not just for stent makers, but also for other device manufacturers that must perform such optical inspections.
Table I. (click to enlarge) A comparison of stent visual inspection approaches.
One of the most critical potential defects is a sharp edge on the inner diameter (ID) of a balloon-expandable stent. If not caught by an inspector, this defect could rip a balloon and cause severe complications during the surgical procedure. Because these extremely fine defects are difficult to spot with human inspection, partly because of saccadic masking, there is always a risk that a small percentage of defects might go undetected. Newer cameras have been developed to meet the increasing need for full-spectrum inspection as stents have become smaller and more intricate. Borrowing from other technologies and modifying optics, lighting, and lenses, a new breed of very-high-resolution line-scan camera systems is becoming the alternative to manual inspection. These cameras generate a large, highly detailed image of the entire stent pattern in a few seconds (typically 5–10 seconds) with no quilting required. Table I compares stent inspection techniques.
Advances in this technology combined with the growth in processing power of computers has enabled the creation of a visual inspection set of tools that is now fast enough to handle even the smallest and most-intricate designs and produce large, sharp images. These tools not only make all the measurements quickly, but they also compare the stent with the original computer-aided design (CAD) model to locate potential visual defects. This combination of technologies was simply not feasible even five years ago.
Breaking It Down
It must be noted that each part of a vision inspection system is unique in its own realm and yet equally codependent on other parts for accurate performance.
Camera and Optics. On the low-resolution end of the scale, line-scan cameras are used to scan postal objects; on the high-resolution end, they are used for silicon wafer inspection. Each camera is developed with special lenses dedicated to the particular application. The latest line-scan cameras typically run from 6000 to 8000 (or even 12,000) pixels in imaging detail and use a 7- or 5-µm square photo-diode sensor. The relatively limited selection of off-the-shelf lenses for these cameras required unique lenses to be custom developed by inspection system manufacturers. The challenge here is that line cameras have a photodiode array approaching 60 mm in length while standard video camera charge-coupled devices are only approximately 6 × 8 mm in size.
To depict stent defects, the camera looks at pixel sizes in the field of view of about 1.5 to 3 µm. Therefore, a lens must work at relatively low magnification (2× to 5×) and project a large (56- mm) image to the line camera. A typical microscope usually has the opposite requirement: imaging much smaller fields at much higher magnifications.
One disadvantage of using line cameras is that since they only acquire one line of the image at a time, they have a very short exposure time. As a result, they require significantly more light intensity than area cameras.
Defect review inspection uses a 360°-rotating mirror assembly below a 2-megapixel area camera. When the system sees something that appears out of the ordinary on the primary or profile view, it drives the mandrel under the rotating area camera to examine the stent in any angle of the 360°.
Lighting. Lighting is always the make-or-break proposition with difficult machine vision applications. Stents act as mirrors because of the high degree of electropolishing that they undergo. Under a traditional microscope, glares or hot spots can appear in the image when the lighting hits a critical angle and all of the light is reflected directly into the camera. Therefore, a critical part of enhancing the camera’s vision required the development of an appropriate illumination system. A real-world understanding of this is easily illustrated: If a person has a flashlight in a dark room containing a wall mirror and the person shines the flashlight at the mirror at just the right angle, it can be blinding. However, at any other angle the mirror looks dark with the light bouncing off the mirror and lighting up the back wall.
The surface image requires a highly uniform illumination so that defects can be seen in good contrast. One solution is an illumination system that physically matches the shape of a stent. It uses a long, narrow beam splitter illuminated by a long, narrow fiber-optic line. A plastic diffuser between the light line and the beam splitter helps to create an evenly lit image of the stent surface with no hot spots present when the camera views the stent surface through the beam splitter.
Many of the stents used to treat medical conditions in the periphery, such as renal or biliary constrictions or abdominal aortic aneurysms, have large diameters that range from 5 to 35 mm. These stents present their own challenges for automated inspection. The primary difficulty is that they are typically made from nitinol and are expanded with heat-based shape setting. This process can cause extensive twisting of the struts—so much so that the stent is no longer purely cylindrical. Consequently, lighting can be even more of a challenge and requires a higher degree of manual verification, thereby slowing throughput. The other obstacle posed by larger stents is the sheer volume of pixels to be processed, which tends to slow inspection and decrease throughput.
Inspection systems can be fully automated once an operator loads the stent and sets inspection parameters.
Software. The setup of stent vision inspection systems is referred to as teaching. System programming, by contrast, is what the original system manufacturer does. Original programming is usually done using C++ instruction code and lower-level drivers. The semiconductor industry often refers to automating an individual part inspection as recipe creation. In the medical device industry, it is alternately called system training, application programming, or part programming. Recipe creation does not generally use computer code but rather involves setting program steps and algorithm parameters using Windows-type click-and-drag features simple enough for the end-user to accomplish.
To make recipe setup easier, system software is needed that can define features as patterns and can search for them within the image recorded by the camera. To create full inspection routines, it may be necessary to attach measurement and defect analysis tools to these patterns. Manufacturers often need to customize defect-finding algorithms by adjusting a variety of parameters to match their specific work instructions and defect definitions.
To support stent development, different vision algorithms had to be developed that could find defects such as sharp edges, mouse bites, or inconsistent laser cuts. The implication is that any variations in the laser path can be found and reviewed. Some systems can use CAD models. They can generate a bitmap image of the ideal part and then compare it with the actual part. Although CAD comparison techniques have been used for some time in quality analysis, such techniques face a unique challenge for stents because these devices are flexible. To compensate for this flexibility, such software must identify key points—typically called warp points—along the curve of the actual stent and then shift them toward the ideal model. This process warps (or merges) the measured and CAD data to each other. The goal is to match the original model and digital photo image precisely, making finer manufacturing defects or deviations from the original design visible.
A good software system will generate a report showing all failed defects and measurements for operator review. It is essential to have operators view the results, because stents are too valuable to discard based on a single alarm from a vision system. The key goal should be to improve quality assurance, though reduced labor and throughput are also benefits.
Once the camera and lighting are set and the software is programmed, the stent is placed on a translucent mandrel usually made of sapphire. To inspect the outer diameter (OD), the stent slides onto this mandrel and light passes from the bottom of the mandrel through the stent as it is rotated 360°. The camera takes a profile image of the part as it spins underneath. The most clinically significant types of defects are best spotted in this view: broken or cracked struts, a bad laser cut, or uneven electropolish. A second scan is then made using the surface illuminator to find scratches, pits, and other cosmetic blemishes. Depending on how the system has been programmed, the software attaches multiple gauges or checkpoints to the digital image and takes measurements or finds defects. A third view is needed for the ID of the stent.
Technologies have been developed to address different aspects of OD stent inspection, but only recently have these coalesced into a robust system that can be used in production environments. The same technology can be applied to ID vision inspection.
Scratches and blemishes can occur on the ID of a stent, and current methods to catch these defects still enlist operators and stereomicroscopes set at about 40× to 60×. The stent is placed on two plastic rods in a hot dog roller– type configuration. If the magnification is high enough and the depth of focus is shallow enough, the operator can focus solely on the inside of the stent, with the outer portion out of focus. The problem is that the human eye tends to tire more quickly when required to constantly refocus on objects that move in and out of focus.
To add more utility to automated line-scan technology, developers have incorporated specially designed optics that support the line camera with 1-µm optical resolution and a depth of focus of about 15 µm. These optics remove the top of the stent from focus and give a clear and accurate digital image of only the ID of the stent.
Instead of using a roller device as in manual inspection, the stent is put on a slotted tube and rotated. The camera looks through the slot with its shallow-depth-of-focus lens and delivers a true image of the ID. Algorithms are applied to the ID picture for inspection in the same way they are applied for OD inspection. Other than the operator loading and unloading stents, the inspection process is fully automated once programmed, and all inspection data are saved within the system software for easy transfer to any PC. If the entire inspection cycle time is less than 30 seconds, then an automated loading system is preferred to maximize efficiencies and minimize the risk of manual errors.
Keeping It Clean
Cleanliness of the stent during inspection is critical to speeding up throughput and reducing false alarms. For example, when a stent is in profile view and it is dark against a bright illuminated mandrel, a speck of dust on the periphery of the stent strut can look like an excess metal defect. It is important to keep the stent particularly clean and dust free from the electropolish line to the inspection station.
In addition to in-house requirements, stent manufacturers must deal with FDA. Stent manufacturers must comply with all FDA regulations, and so must the inspection equipment. In addition, the equipment must undergo installation qualification—all electrical and mechanical setups are checked and certified to make sure they function properly. Then the inspection process is put through operational qualification, where the system is tested under worst-case conditions. Finally, qualification is performed under normal conditions according to GMP.
The vision inspection system has to be certified and validated before any specific stent design can be processed. Attaining this qualification and completing the associated paperwork is often the biggest task in preparing the inspection system for use.
Implementing a system to find critical defects is a more significant undertaking than validating a system to do dimensional measurement. For dimensional measurement, an engineer can verify that each feature of the software functions properly. Then the system can be run through a gauge repeatability and reproducibility test, in which different operators repeatedly measure a statistically meaningful set of sample parts. All the data can then be brought down to a percentage of manufacturing tolerance used up by the measurement instrument.
Validation work is generally broken down into two broad categories: system validation and recipe validation. System validation comprises the installation and operational qualification (IQ/OQ) and the process qualification (PQ). The IQ and OQ merely validate that the required software and hardware components are present and fully functional. The PQ is much more involved; for a given stent or stent family, the inspection process must be proven to be effective, and any risks in its use must be understood and mitigated. Each algorithm used must be understood and challenged.
For validation of visual inspection, how can the stent maker be sure that some new process variation might not produce a defect the system has not been programmed to find? A number of strategies can mitigate this risk. All strategies stem from the basic concept of understanding the process, finding its weakest links, and exhaustively challenging those vulnerabilities.
The first strategy is to assemble a physical library of defective stents that contains multiple examples of clearly rejectable, barely rejectable, and barely acceptable defects for each classification. Can the system create an image that first allows a human to look at the screen and see the continuum of defect severity? Can the system then be tuned to consistently discern the barely rejectable from the barely passable? If the system can do these tasks, an engineer can infer that the more-critical clearly rejectable defects will definitely be captured.
The next strategy is to understand the algorithmic approaches taken by the instrument maker. Review images of actual defects and then use a program such as MS-Paint or Adobe Photoshop to create subtle, artificial defects in the image. For example, one might take an image of a perfectly good stent and draw in a well-formed strut that does not belong there to see whether the system can find it.
The key here is that validation engineers must understand how the system works down to the algorithmic level to effectively come up with what-if scenarios that will expose the system’s vulnerabilities. Another consideration for validation engineers is the degree of sensitivity. While every anomaly the system finds on a given part is shown on a screen for review by an operator, it can be time-consuming to review them all if the algorithms are tuned to be too sensitive to anomalies.
Each new family of stent designs requiring a new recipe should not trigger a PQ from scratch. Validation engineers must determine the following: Do the defects on a particular stent appear on the image in the same locations as similar defects in previously validated products? If there is any fundamental change to the production process (e.g., a longer electropolish) that leads to a more rounded strut edge, the PQ for this family of parts must be nearly as extensive as the initial one.
Adapting the Technology
In addition to stents, many other devices in the medical industry are cylindrical. These too are shrinking in size. Accurate line-scan vision inspection can be used for precision tube products, connectors, catheters, guidewires, needles, angioplasty balloons, and stent grafts as well as stents.
In the world of flat products, line scanning works well as long as there aren’t too many step-heights to the product that would require many extra scans. A tiny gear with two or three different heights might still be a good candidate since it would take minimal time or programming to run three different scans. However, when dealing with prismatic devices or those with a wide variety of heights, a more traditional video camera mounted on an x-y-z stage is a much better method.
The basic platform for this type of line-scan vision inspection allows easy adaptation to a wide variety of production needs that might in other cases require a full-fledged R&D effort to develop. For programming individual parts with multiple repetitive features, pattern-find features often allow these types of systems to search the entire image and find each instance of a feature, even if there is no particular order or pattern to the feature locations. Finding these features shortens time to implement new designs or process improvements with complete data. It improves the production process through accurate process control data. In the area of stent inspection, it reduces the overall cost to inspect on a per-stent basis while improving reliability through automation.
Another area suited to this technology is the inspection of mass quantities of very small parts. Such parts can be machined, laser cut, molded, or photo-etched. Once parts like these are below 0.2 in. in size, they become difficult to fixture. The pattern-matching or scanning technology described here can image these parts and use pattern matching to find and orient each individual part and perform quality analysis.
Changing from a dedicated team of 50–60 people using microscopes to an automated method may often require investing in a custom vision system. The transition is made easier with image processing, defect detection algorithms, and high-powered cameras. Stent makers and other medical device manufacturers can now invest in digital line-scan systems that are dedicated to this type of inspection. Most systems are 95% ready to go, requiring changes only for individual production needs. Therefore, the task of automating what used to be a labor-intensive product inspection process becomes more of an applications engineering task rather than a full-scale development project.
Manufacturers of stents, whether they are OEMs or contract manufacturers, can benefit greatly from this new technology. The future will tell whether device manufacturers will embrace the technology and modify it to suit their particular inspection needs. The technology continues to evolve and take advantage of the newest developments in the growing field of digital optics, imaging, and computer capabilities.
Dan Freifeld is president of Visicon Inspection Technologies in Napa, CA. He can be contacted at email@example.com.