Barcode scanning is of utmost importance to research and medical laboratories alike. Specimens that fail to scan can cost valuable time and potentially waste irreplaceable specimens, resulting in patient harm and/or sunk costs. Barcodes are used in a variety of other industries as well, such as the automotive sector and construction, which rely on them with sometimes similar urgency; it is here that some have turned to AI to increase the accuracy and consistency of scanners to read barcodes.
How scanners read barcodes
Most typical barcode scanners, like pen-type and laser scanners, utilize a light source to scan barcodes. They register light intensities as they bounce from the scanner to the barcode and back. Waveforms are generated that register the width of the bars and spaces, which are then decoded. Some scanners, like charge-coupled device scanners and camera-based scanners, use ambient light instead of producing a laser to register bar widths.
Issues with barcode scanning typically arise due to interference from a variety of sources. The most prominent sources of interference include bleeding of ink of the barcode or when abrasion occurs, rendering the barcode much more difficult to read. While these problems can often be mitigated using thermal-transfer printing methods, they are difficult to fix once they occur, as scanners are unlikely to decode an improperly generated or damaged barcode.
Adapting AI to scanning, processing, and automation
Where barcode scanners are used, a high volume of samples is likely being read, making it difficult to examine where a barcode reading error may occur and why. Thus, companies like Cameralyze and Leuze have developed AI-based systems that integrate with barcode scanners to help identify, track, and troubleshoot barcode errors. Leuze, in particular, has worked within the automobile sector to develop an algorithm that assesses the frequency of barcode errors and attempts to discern whether they occur more frequently with a specific type of label, reader, and/or installation. To perform the analysis, it utilizes machine learning to generate an algorithm similar to that used by streaming services, which analyze user behavior and recommend films. Here, the algorithm rates the barcode label based on its fit with a specific reader, like it would match a viewer based on their fit with a given film.
These types of AI-based systems can be implemented using a cloud service or edge device. An edge device is a distinct device, separate from the installation, that obtains data, performs analyses, and passes the results to other platforms. It can also communicate with the barcode reader itself, which allows the reader to report errors as well. Edge devices and cloud-based systems are recommended because they do not require altering current IT architectures.
Adapting AI to barcodes in science and healthcare
The impact of AI-based barcode reading algorithms in science may ultimately prove immense for labs that process hundreds to thousands of samples daily. Patient samples, in particular, should carry as little risk as possible when it comes to barcode scanning accuracy and consistency, as even a single misread barcode can have negative consequences, such as wasted tissue specimens and false test results. Therefore, the impact of AI-enhanced barcode reading may one day be felt across healthcare, should hospitals and medical laboratories decide to eventually implement it.
In research, AI-based barcode reading has already been used at least once. Though the scientists did not utilize AI specifically to enhance the reading of damaged or poorly visible barcodes, it was implemented with a barcode-based system to monitor honeybee behavior. Here, barcodes were stuck to honeybees, and a convolutional neural network was adapted to identify specific behaviors and confirm their relevance.1
LabTAG by GA International is a leading manufacturer of high-performance specialty labels and a supplier of identification solutions used in research and medical labs as well as healthcare institutions.
References:
- Gernat T, et al. Automated monitoring of honey bees with barcodes and artificial intelligence reveals two distinct social networks from a single affiliative behavior. Sci Rep. 2023;13:1541.
