Solutions are underway to address challenges in code reading, substrates, lighting, validation, and more.
By Rick Roszkowski, Senior Director of Marketing, Vision Products Business Unit, Cognex Corp.
FDA’s final rule establishing a Unique Device Identification (UDI) system requires the device labeler, in most cases the manufacturer, to include an automatic identification (AutoID) code on device labels and packages and, in the future in some cases, on the devices themselves. In most cases, complying with these regulations will require labeling packages and eventually products with 1-D or 2-D codes as well as a human-readable code with batch-level identification. These codes nearly always need to be verified at the time they are marked or printed and at additional times downstream in the process. Code reading can be challenging because of the potential for code degradation and the wide range of potential surfaces that may need to be marked. Another potential challenge is that vision systems need to be validated based on Good Automated Manufacturing Practice (GAMP) in order to provide documented assurance that they are appropriate for their intended use before going live.
The machine vision industry is helping to address these challenges by developing software algorithms that are able to find and decode even damaged 1-D or 2-D codes and read low-contrast or damaged text. Another advancement involves application-specific solutions for vision systems that eliminate the need for programming and prebuilt documentation that substantially reduces the time required for FDA validation.
FDA issued the rule in 2013 requiring that the medical device industry must include the UDI as both a bar code and in human-readable text on device labels and packages. The UDI must include a device identifier (DI) that identifies the labeler and the specific version or model of device and a production identifier (PI) that includes or more of the following: lot or batch number, serial number of a specific device, expiration date of the device, and the date the device was
FDA categorizes medical devices into one of three classes—Class I, II, or III—based on their risks and the regulatory controls necessary to provide a reasonable assurance of safety and effectiveness. Class I devices generally pose the lowest risk to the patient and/or user and Class III devices pose the highest risk. Class III medical devices have been required to carry UDI labeling since September 2014. Implantable, life-saving, and life-sustaining Class I and II devices are required to carry UDI labeling as of September 2015. All Class II devices must carry UDI labeling and Class III devices that already require direct part marking (DPM) must be marked with the UDI on the device by September 2016. By September 2018, all Class I devices must carry UDI labeling and Class II devices that require DPM must carry UDI on the device. Finally, Class I devices that require DPM must carry a permanent UDI on the device by September 2020.
Smart camera vision systems are a critical component of the UDI labeling process, not only to validate the presence, accuracy, and readability of the various labels, but also to ensure product safety and package integrity. The UDI code is often read many times as the product moves from manufacturer to consumer. In time, all entries will be registered in a global database to document the chain of custody during the entire product cycle, making it possible to easily and quickly verify the authenticity of each pharmaceutical product. While some companies have considered using their existing workforce to manually track and trace the packages, there are concerns that human operators are vulnerable to fatigue, distractions, and interruptions that may detract from their performance and make it impossible to provide 100% assurance of an accurate pedigree. Manual operations are inherently vulnerable to untraceable errors, making it difficult or impossible to comply with UDI regulations.
Code reading advancements
Nearly all medical device manufacturers and other device labelers are using smart camera based vision systems that are controlled by internal microprocessors so they can operate independently of a PC. Vision systems are less expensive to implement because they can typically be developed without writing a line of code by utilizing a few prewritten functions called vision tools.
Operators can adjust the focus or lighting on the vision system either by plugging in a laptop or by operating the vision system in teach mode. This approach makes it easy for the end-users to find “like for like” camera replacements for many years after initial installation and to maintain consistent vision performance across multiple inspection points. The smart camera–based solution provides medical device manufacturers with a lower cost of ownership by reducing the need for revalidation because the vision system is inherently much more stable over time and is not subject to yearly computer obsolescence issues.
Reading UDI 1-D and 2-D codes and text can be challenging in real-world applications because of variations in the size, shape, position, and orientation of the code; the potential for degradation in printing and marking of the codes; the wide range of surfaces on which the codes may be marked; the potential for degradation in the equipment used to print or mark the codes; and variations in ambient lighting. Proper, high-quality marking is essential for full supply-chain readability, but as the initiative moves forward to direct-part marking of individual devices, the challenges are amplified. Fortunately, vision system suppliers have developed solutions for these problems.
A key advancement in 1-D code reading is an algorithm that uses texture to locate bar codes at any orientation and then extracts high-resolution signals for decoding. The finder analyzes a raw source image and produces a list of regions where it is likely that a code exists along with the orientation and other properties of the code. The algorithm then extracts 1-D signal using as a mathematical foundation a model of the pixel grid itself that reduces blur while maintaining perfect accuracy and noise reduction.
Also, 2-D code reading improvements include new texture-based location algorithms that take a unique, inside-out approach to reading 2-D matrix and DPM codes. While conventional feature-based algorithms start by locating the finder pattern, the new algorithm looks for a pattern of alternating light and dark modules within the code. This technology dramatically increases read rates in 2-D bar code reading applications where a part’s geometry, poor lighting, occlusion, or print-registration errors make it difficult to capture an image of the entire code. Unlike previous solutions, new technology can locate and read codes even when they exhibit significant damage to or complete elimination of the finder pattern, clocking pattern, or quiet zone.
Reading human-readable text
Similar improvements have been achieved in the even more challenging task of reading the human readable text that is required to meet UDI regulations. The new optical character recognition (OCR) technology is based on a powerful image processing algorithms wrapped in an easy-to-use point-and-click interface. The segmentation process first breaks down each line, using regions around each string, and automatically locates them. Next, segmentation parameters are adjusted in order to get boxes around each character. Each character is broken down further into small fragments and the segmenter finds each letter based on skew, min/max height, min/max width, minimum aspect ratios, angle/skew, and other characteristics that define the font. Finally the characters are trained so they can be recognized. To improve read rates, the segmentation tool also provides a noise filter, stroke width filter, compensation for changing lighting conditions, and automatic character scaling. This all sounds quite
complicated, but with auto-tune capability, operator input is minimized to placing a box around each string and entering in the characters. The auto-tune function then dramatically decreases the time it takes to set up the tool by acquiring a sample image and automatically adjusting the tool to its optimal segmentation parameters which otherwise would be done manually. While finding the segmentation parameters auto-tune also trains the characters at the same time.
Vision system suppliers have also made efforts to simplify the process of implementing vision systems for UDI applications and validating them to GAMP standards. For example, an out-of-the-box solution for UDI applications can read both 1-D and 2-D codes and text and can automatically validate bar code information against standard GS1 Application Identifiers. No programming is required. The application is configured by following step-by-step menus to set up the job parameters. On-line grading is used as a feedback tool to tell you whether print quality has changed, which could mean it’s time for preventative maintenance such as cleaning the print heads on a printer.
An audit server can also be used to track significant events. The application includes a password-
protected user account control, with parameter access control for different user groups. An audit server runs on a PC and tracks significant events on the camera. When the audit server is enabled, cameras send XML formatted messages to the audit server application whenever a user logs in, changes a job, puts the camera on-line or off-line, and changes a parameter. If the audit server is down, the camera will buffer up to 1000 events and then transmit them after the connection to the server is restored. This approach meets the requirements established by the FDA’s 21 CFR Part 11 regulations for electronic signatures and records.
IQ/OQ validation documentation
The use of a pre-written job template minimizes the amount of work required to validate the line. Templates are provided for the Functional Specification (FS), Design Specification (DS), Installation Qualification (IQ), Operational Qualification (OQ), and a corresponding Trace Matrix document that ties these documents together as required by GAMP V validation procedures by correlating the FS and DS to the IQ/OQ documents.
Recent technology advancements are benefitting medical device manufacturers and other supply chain contributors rushing to comply with UDI regulations. For example, in the vision-based ID reader space, software algorithms have been developed that are able to find and decode even damaged and poorly marked 1-D or 2-D codes through a wide range of illumination, marking, and material variations. OCR software has been developed that can read text even when there is little contrast between type and background and when letters are touching, skewed, and distorted. These advancements will help position the medical device supply chain to achieve compliance with current and future regulations, reduce costs, and improve the security of their supply chain.