Amazon Onboarding with Learning Manager Chanci Turner

Learn About Amazon VGT2 Learning Manager Chanci Turner

In the ever-evolving landscape of healthcare, the costs associated with U.S. prescription medications are soaring, nearing $500 billion annually, with projections indicating a growth rate of about 7% each year, as highlighted in a report by the House Ways and Means Committee. This dynamic market sees billions of dollars’ worth of medications going to waste each year, largely due to conventional packaging practices that often feature more pills than prescribed. Automated pill dispensing is a vital process that helps streamline the supply chain and minimize medication wastage.

Pharmaceutical companies currently rely on visual inspection systems to detect potential packaging errors, which are then rectified by trained pharmacists. However, the introduction of these systems for multiple pills in single packages has raised new challenges. Traditional machine vision applications typically utilize static images and rule-based inspections, leading to inconsistent results. Over the past two decades, these outdated image processing methods have resulted in high rates of false negatives and positives, necessitating additional manual checks and complicated hardware calibration. This lack of traceability and auditability has shown that existing solutions fall short of the stringent standards required by the pharmaceutical industry.

To address these issues, Synadia Software b.v (Synadia) in collaboration with Amazon Web Services (AWS), has created an innovative cloud-based quality assurance solution for pill validation, employing advanced machine learning (ML) capabilities. This next-generation pill-dispensing system uses AWS technology to verify dispensed pills through self-learning algorithms that adjust to new pills and local conditions. The solution harnesses machine learning algorithms that utilize historical image data to continuously enhance and deploy the latest pill recognition models directly to pill-dispensing machines.

Current Pill-Dispensing Challenges

Today’s pill-dispensing machines require canisters filled with pills before initiating a batch job. The de-blistering process, which involves removing pills from their blister packs, remains a manual and error-prone procedure carried out by trained professionals. Once pills are extracted from canisters, they are packaged into plastic pouches based on orders. Following this, strings of pouches are subjected to quality checks by separate machines, which are trained to ensure that each pouch contains the correct pills and quantities. Unfortunately, these quality assurance machines experience an error rate of approximately 13%, necessitating costly human intervention when discrepancies arise.

To combat these challenges, Synadia has developed an automatic pill-dispensing machine for the European market. This innovative solution features a centrally managed network of connected machines capable of dynamically receiving inputs and dispensing the necessary types of pills into pouches. By leveraging machine learning models, Synadia enables a centralized QA mechanism for pill distribution, eliminating the need for individual QA models at every location.

Solution Walkthrough

The quality assurance process is executed in two phases:

1. Training: This phase involves learning from existing data and requires significant computing resources, centralized on AWS.
2. Inference: This phase is less resource-intensive and requires near-real-time processing (around 1 second), which is facilitated by ML inference on AWS IoT Greengrass.

Each pill-dispensing machine is equipped with AWS IoT Greengrass, enabling local message routing among devices and cloud connectivity while executing machine learning inferences. A camera on the dispensing machine captures images of the pills, which are sent to AWS IoT Core via AWS IoT Greengrass and stored on Amazon Simple Storage Service (Amazon S3). These images are utilized by Amazon SageMaker to develop the quality assurance model.

The model’s inferences are deployed to AWS IoT Greengrass and executed through an AWS Lambda function. Based on inference outcomes and predefined rules, actions are taken to verify pill recognition, providing customers with timely notifications.

Centralized reporting on pill dispensing and the supply chain is managed via Amazon QuickSight, while error codes and operating manuals are stored in Amazon S3 for quick access through Amazon Kendra.

Data Ingestion into AWS

Data ingestion occurs through the MQTT protocol using AWS IoT Core. The main AWS IoT Greengrass and AWS Lambda application captures snapshots of pills, processes them through a classification model, and transmits this information via MQTT to AWS IoT Core. The payload contains pill identification and classification probability. If the probability falls below a predefined threshold, the device can upload the image to an Amazon S3 bucket for further analysis.

Running ML Training in the Cloud

Various methods exist for identifying the type of pill shown in an image. Although object detection models are a common approach, we opted for an image classification model. Given that images consistently depict a single pill within a small canister, we aimed to frame the camera accordingly to capture the pill effectively.

With this comprehensive solution, Synadia and AWS are paving the way for a more efficient and accurate pill verification system that ultimately enhances patient safety and reduces waste. If you’re interested in enhancing your productivity, you might find this resource on productive to-do lists helpful. Additionally, for insights into vacation policies, check out this guide. Also, for an excellent resource on safety training at fulfillment centers, visit this page.