Key Figures from our use case deployment
Our model delivered an accuracy of 98% in precise detection of pills, 96% in detection of extraneous materials or debris
Automating pill detection resulted in 45% higher efficiency, lower cost per package, allowing scalability of adherence packaging with rising demand
A staggering 55-60 % of patients worldwide suffer from medication non adherence. The severity of non adherence affects the overall health outcome of patients, but also drug manufacturers, pharmacies or even healthcare payers suffer losses estimated at $600 Billion annually!
In adherence packaging, drugs are packed in individual pouches according to the dose, date and time of consumption. In such cases, stringent quality check to ensure right medications are placed in right pouches is critical for the success of adherence packaging.
Data Science Methods Used
Finarb’s AI pill detection platform is at the heart of this automated quality check and is based on Computer Vision - precise object recognition algorithm to identify, pre-categorize and validate contents of a drug pouch to ensure customers receive the correct drug product based on several drug attributes as well as the National Drug Code (NDC).
The workflow involves using pill images and configuration data from the adherence packaging machines, sending that data to blob storage, triggering training of the AI model to identify set features/metadata/NDC of the pills, and then running a batch inference on a set of images. This enabled an expedited and highly accurate pill detection process.
Benefits of using AI in pill detection in medication adherence packaging
Accuracy of our AI model in pill detection
The model after 2 iterations and pilot testing delivered an accuracy of 98% in precise detection of pills, 96% in detection of extraneous materials or debris
Additionally our ensemble model combined results from 6 individual AI computer vision models for higher precision
The model is now being trained to identify larger set of NDC’s with integrated image augmentation technology to increase the accuracy and precision of object identification and categorization to close to 99%. Through distributed computing, we have also expedited the AI training and processing time to scan these large datasets
Automating this check has resulted in 45% higher efficiency, lower cost per package, allowing scalability of adherence packaging with rising demand
Finarb and our client have filed a patent for this pioneering workflow leading to higher accuracy in pill detection Vs benchmark