Building an AI/ML application for the Medical Industry using Kubernetes
Hybrid Cloud Patterns: What Are They?
Hybrid Cloud Patterns are a natural evolution from reference designs, providing added value for Building an AI/ML application for the Medical Industry using Kubernetes
Customer solutions involving various Red Hat products are the focus of this initiative. One or more apps based on successfully deployed client samples are included in the patterns. As an example, a sample application code is provided, as well as the numerous open-source projects and Red Hat products required to make the deployment work. The pattern can then be tweaked to fit the needs of the user Kubernetes.
How do we choose and make a pattern? We seek innovative client use cases, obtain an open-source illustration of the use case, validate the pattern and its components with appropriate product engineering teams, then automate them using GitOps to make them readily repeatable and expandable.
The automation also enables the solution to be added to Continuous Integration (CI), with triggers for new product versions (including betas), so that we may discover and resolve breakage and avoid bit-rot in advance.
Who should make use of these designs?
These patterns should be used by architects or advanced developers who are familiar with Kubernetes and the Red Hat OpenShift Container Platform. As part of the pattern framework, advanced Cloud Native concepts and projects are delivered. OpenShift Gitops (ArgoCD), Advanced Cluster Management (Open Cluster Management), and OpenShift Pipelines are just a few examples (Tekton)
The goal of this pattern is to demonstrate how medical institutions might use trained AI/ML models to detect irregularities in the body, such as pneumonia. From the standpoint of medical workers, it works with medical imaging equipment, requiring the submission of an X-ray image into the application to begin the workflow.
The image is saved to an object storage system that is compatible with S3. This upload causes a storage event, “a new image has been uploaded,” to be broadcast to a Kafka topic. A native Eventing listener consumes this topic and launches a KNative Serving instance as a result. This instance is a container image that contains the AI/ML model as well as the processing functions required. The container fetches the image from the object store using the information from the event is received, pre-processes it, uses the AI/ML model to forecast the risk of pneumonia, and records the result. The medical personnel is also notified of the results.
Implementation of Patterns
Follow the instructions on the getting-started page to deploy this pattern.
What’s going on?
The initial openshift-gitops operator is provided with the requisite custom resource definitions and custom resources to deploy the datacenter-validated-pattern> with references to the relevant git repository and branch during the bootstrapping of the pattern. The argoCD application will construct all of the common resources, such as advanced cluster manager, vault, and openshift-gitops, once it is deployed. The pattern deployment starts with argo applying the helm templates to the cluster, which leads to all resources being deployed and the xraylab dashboard being accessible via its route.
The pattern deployment charts may be found under $GIT_REPO_DIR/charts/datacenter/
Technology with a Pattern
Because of the critical reliance on the original information, several resources did not align with 1:1 and had to be overcome. For example, there are several activities that search the cluster for information to convert into a variable and then apply that variable to a resource. As you may guess, stating the status of your cluster can be really difficult. To get around these mandatory activities, we did what we could and used OpenShift jobs to complete the work.
The complete pattern can be found here. When evaluating the benefits of this pattern, words like speed, accuracy, and efficiency spring to mind. Because we can employ technology to swiftly and reliably diagnose anomalies discovered in X-rays, patients get the treatment they need when they need it. Administrators may easily answer user expectations using the framework of the verified pattern by delivering solutions that just require users to contribute their own data to complete the last 20-25 percent of the design.