The label is a description of a drug product or a device provided by the manufacturer and approved by the regulatory authority of a particular country or jurisdiction. This includes indications for its use, who should use it, adverse events, instructions for use, and safety information. Labels seek to identify contents and the specific instructions or warnings for administration, storage, and disposal.
Labeling is the most important part for all types of products across domains, such as medical, home appliances, drugs, automobiles, etc. Each product may contain 5 or 6 labels, including product labels, carton labels, and some more.
Labeling requirement situation
Each product has its own regularity rules with respect to its domain. As per the technological update, these regulations get updated every decade or less. For the regulatory change, all labels must be compliant. So all the labeling requirements must be updated under the following steps.
- Label gap assessment
- Label redlining
- Label design update
- Label proofreading
Label gap assessment
Labeling gap assessment is nothing but checking the content of labels as per the regulatory compliance. Mainly three types of content need to be verified; symbols, barcodes, and text. Symbols include CE marking, factory symbol, expiry date symbol, and keep dry, etc. The text includes factory address, expiry date, serial number, temperature, reference number, etc. All the regulatory requirements get captured by using the excel sheet called gap assessment sheet. Then, each label is compared with the requirements available in the gap assessment sheet. The gap status is updated in the same gap assessment sheet as fully gap, partially gap, and no gap. Here, the final output is the filled gap assessment sheet with all the gaps.
Label redlining means placing the missed symbols/content in red color on the particular label with respect to the gap assessment summary sheet. The gap assessment summary sheet is the output of the above gap assessment process, and the same is the input for the label redlining process. Depending upon the gaps, there are different types of redlining, such as adding missing symbols, striking and adding the right text, striking and adding the right symbols, adding text comments, etc. Here, the final output is a redlined copy of the existing label.
Label design update
Once the redlining is completed, all the label defects must be updated/corrected as per the redlined comments. This can be done by using the labeling update tools. Here, the output is the clean copy of the label, which is updated with all the redlined comments.
In this step, all the updated label copies must be compared with the redlined copies to make sure all the comments are updated in the clean copies. If any comments are missed, the clean copy is sent back to the previous label updating stage to correct the same. Once again, proofreading is done.
Currently, the entire labeling process is manual, and it consumes considerable effort. Each label takes manual effort, each product contains 50 to 60 labels, and each product manufacturer has 100’s of products, so the overall time consumption is high.
The labeling process can be automated using current technologies, such as computer vision, OCR, image processing, etc.
Object detection is a computer technology related to computer vision and image processing that deals with detecting instances of semantic objects of a certain class in digital images. Here, we can use object detection for detecting symbols availability and getting location information within the labels. This can be achieved by using neural network deep learning models like convolutional neural networks (CNNs). Initially, this deep learning model requires training with all the symbols, and it is named as different classes. Once the label is passed as an input to the object detection module, it will apply the convolution technique along with the neural network to find the symbols and the location coordinates as a result. Using this result, we can compare it with the label compliance requirement.
OCR (optical character recognition) is a technology that recognizes text within a digital image. It is commonly used to recognize text in scanned documents. All the label text content verification can be done using this module, which will extract the text from the label image and compare it with the reference text. This is available in the requirement sheet.
This automation solution will reduce 60% of manual effort during the labeling gap assessment and its redlining.
- Deep learning training is required whenever a new symbol is added to the requirement, and it takes more time to do the same. This can be managed by using the transfer learning technique.
- The quality of the label will impact the output accuracy. This can be addressed by using image pre-processing techniques, such as image gray scaling,pixel brightness transformations/brightness corrections,image filtering and segmentation, etc.
- Object detection requires high computation power, and it takes more time to generate the output. This can be managed by using the GPU-enabled PC.
- The deep learning model requires training with lots of data to improve accuracy. Image augmentation techniques can help overcome this.