IBM Custom Connectors for NHS.

At first when we were introduced to our scenario, we thought of two different potential software use cases that utilized IBM Cloud APIs. They are:

• AI medical data analysist - For each returning patient, an AI data analysist can search up all of his/her past medical records, analyse it and suggest several potential health issues and their respective treatments for the patient with risk estimation to the doctor before his judgement. This aims to reduce the time for diagnosis, thus to reduce the chance of missing the golden opportunity to treat the patient before it is too late. Patients with late-stage cancer and organ failure may greatly benefit from it.


• Doctor appointment assistant - An online chat bot operating 24/7 that helps patients make doctor appointments. It may also help suggesting patients with simple self-aid treatments if there was a minor illness/injury. This aims to reduce the repetitive and quantitative workload of nurses, so that they could focus more on qualitative work such as comforting patients in a hospital. This also aims to reduce the reliant of patients with minor illness, so that clinics and hospitals can spare up more capacity for patients with more serious and urgent illness.

With the mentioned two use cases in mind, we narrow our choice of APIs down to Watson Discovery, Watson Assistant and Annotator for Clinical Data.

• Watson Discovery¹ - an intelligent document processing engine that helps discovering valuable data within long and complex documents. This can help identifying and extract illness related keywords such as symptoms for easier diagnosis.


• Watson Assistant² - cross device, application or channel chatbot equip with self-learning AI capabilities to understand questions that customers ask in natural language. This can help appointment making experience more personalized and convenient.


• Annotator for Clinical Data³ - extracts clinical insights, medical concepts, and medical codes from unstructured data with pre-built but highly customizable annotators and cartridges. This is Watson Discovery specialized in the medical field.

While the use of AI carrying out pre-diagnosis and suggesting suitable treatment may sounds cost effective and possibly even with a higher judgement accuracy than doctors, it is controversial as an unethical way to treat patients, especially those with their lives depending on it. We therefore decide not to include this function into our use cases.

After some research and experiment, we found out that Annotator for Clinical Data API under the scope of Merative⁴, formerly IBM Watson Health, was no longer available in IBM Cloud a few months after this project began. The accessibility of Watson Discovery on the other hand, was locked behind a monthly subscription for its strong capability which our IBM client could not grant access to students. Watson Assistant therefore became our final go to API for this project.

To enable the use of IBM Watson APIs in Microsoft Power Platform, a custom connector can be built with a single API using either Power Automate or Power Apps. There is no difference in how custom connector behaves across the two application and building a connector on either one of them can be reused by the other. The process of building a custom connector can be divided into three different steps. They are security setting, definition setting, and testing.

Each connector requires an Authentication method based on how the used API specification. The connector is only said to be functional when the correct authentication is passed to the API, which grant access to sending and receiving messages between the connector and the API. For all IBM Watson's API, a basic authentication is used with the field 'username' filled with the fixed string “apikey”, and the field 'password' filled with the unique API key given in the user's IBM Cloud account. The other available authentication methods that is not used by the IBM APIs include API keys and OAuth2.0.

The definition of a connector represents its triggers and actions, meaning how the connector will behave when a certain command is received. This is similar to a method in coding, which also refers to the methods in an API. Each method offered by the API should be built into one action in a connector using an URL. While the best solution is to include every single method given by the API into its connector, we are aware that our limited time on the project is not enough for a fully built connector, thus we have decided to only include the methods that most suit the environment of building an app using Power App. For example, in our Watson Assistant connector, we only included 'create session', 'delete session', 'send user input to assistant (stateful)' and 'send user input to assistant (stateless)' methods which allows users of a mobile app using the connector to connect and communicate to Watson Assistant. Without other backend methods like 'list log events' and 'update environment', which only the developers should use and users should not have access to.

While building actions and triggers can be performed using the template provided by Power Apps, custom code can be used on each action to further advance its functionality. As in our project we aims to provide a general solution for the public to expand on, rather than a connector specialized in performing a certain action, we built our actions without any custom codes.

Testing a custom connector can be done by performing actions and sending request to the API and checking if a correct JSON response is received. It can also be done by test running with a canvas app, or a flow template. This will be further explained in the 'testing' section.

While connectors can be built from scratch by going through the steps mentioned above, there is also other workarounds such as importing an OpenAPI specification⁵ (Swagger specification) or a Postman collection⁶. Importing by either methods will automatically generate all actions and codes listed inside the imported file. This is useful especially when we tried to understand working and testing with Watson APIs at first using Postman, without operating on Power Platform, and we wish to save some time and reduce human error when implementing it.

---

References

1. Watson Discovery
2. Watson Assistant
3. Annotator for Clinical Data
4. Merative
5. OpenAPI Specification Github
6. Postman collection

---