Difference between revisions of "Cyclops"

From wiki
Jump to navigation Jump to search
m (→‎Repository: Added links to repos)
 
(3 intermediate revisions by 2 users not shown)
Line 1: Line 1:
=== Stream I ===
+
The [[Technology and Business Innovation]] (TBI) under the [[Planning and Operations Directorate]] (POD) undertook Project Cyclops funded by the [https://www.gcpedia.gc.ca/wiki/Health_Canada_Innovation_and_Experimentation Solutions Fund] to bring innovation within [[Health Canada]]. Although not a one-eyed superhuman strength mythological creature, Project Cyclops aims to develop an online web application that would enable inspectors to use a digital tool/camera to upload images to assess whether a product label is compliant with Health Canada’s regulations.
The project entered Stream 1 in September 2018.
+
 
=== Stream II ===
+
The scope of the project was limited to focus on [https://www.canada.ca/en/health-canada/services/drugs-health-products/natural-non-prescription.html natural health products] (NHP), in partnership with Inspectors from [[Health Product Compliance and Border Integrity Unit]] within [[ROEB-Digital Transformation|ROEB]], and financial and technical support from Solutions Fund and IMSD.
The project was approved for the next phase in July 2019.
+
 
=== Pilot ===
+
The current label review process is tireless, tedious and time-consuming. The process can have more than 10 long manual steps. An inspector can spend hours reviewing multiple pieces of information on a label, such as medicinal and non-medicinal ingredients, dosage, and claims and risk information comparing that to the licence issued by [[Health Canada]]. Such inspections are highly repetitive and require very precise focus and attention to details.  
 +
 
 +
=== EXPERIMENTATION ===
 +
The project was separated into two streams:
 +
 
 +
* Steam I was a beginning stage into exploring various options and elements that would be necessary for developing a minimal viable product to be built in order for the solution to deliver the provided result. Steam I first goal was to undertake a feasibility study, identify challenges, options analysis and a Proof of Concept (PoC) model.
 +
* Stream II first focused on the development of machine learning models, following by the development of the solution pipeline that can incorporate the 4 key areas of focus. The solution pipeline includes the privacy design approach to development, rules governed by statutory constraints with inspections enabled by telecommunications, data source integration and automation, and UX design.
 +
 
 +
In Stream 2, a prototype was developed and tested through a partnership with developers for Statistics Canada that would allow inspectors to review labels on-site, by:
 +
 
 +
* Virtualizing optically represented text on labels,
 +
 
 +
* Querying data via existing databases,
 +
* Identifying and highlighting instances of non-compliances,
 +
* Visually depicting any issues by overlaying information back onto the application.
 +
 
 +
=== BENEFITS TO HEALTH CANADA ===
 +
The Inspectors monitor compliance and conduct enforcement actives to prevent the sale of non-compliant products and protect the health and safety of Canadians. The tool built through this project will enable inspectors to review more product labels and refocus their efforts on other parts of the inspection process and help increase confidence in Health Canada’s national compliance and enforcement process. Product testing by Inspectors in [[ROEB-Digital Transformation|ROEB]] informed us that inspectors valued this direction as 80% of testers indicated that a fully functioning prototype would increase confidence in the accuracy and the number of inspections.  
 +
 
 +
While currently only being developed for NHPs, this technology has the potential to scale and be repurposed for inspections across other products regulated by Health Canada, such as consumer products, pesticides, drugs etc. TBI has already started engagements with the Tobacco and Vaping Compliance and Enforcement Program for the development of technology to scan labelling of tobacco products. Furthermore, there is an opportunity to develop a similar product for Canadians to verify health information at points of sale, receive any risk notifications (recalls, advisories etc.) and empower them to make informed decisions on the health products they consume.
 +
 
 +
=== EXPERIMENTATION PHASE ===
 +
Working in the experimentation space, we were able to quickly transform the project from just an idea into a fully developed prototype through rapid approval, disbursement of funds and support from Solutions Fund. We also encountered broad support from Inspectors & management within [[ROEB-Digital Transformation|ROEB]], despite the pandemic, to make the efforts necessary to test and collect data.
 +
 
 +
We recognized how invaluable it was to build partnerships with Industry and other Departments to get the necessary expertise in AI and Machine Learning technologies. Solutions Fund through their various initiatives were able to assist our efforts, and connect us to other departments pursuing similar projects, allowing us to leverage the knowledge from previously conducted. As there are various options in the industry in this highly dynamic and constantly evolving subject, it was important for us to continue researching while doing work in existing phases. The project during its lifecycle shifted 3 times in development from vendors such as [https://lixar.com/ Lixar], [[Statistics Canada (StatCan)|Statistics Canada]] and [http://www.cistel.com/ Cistel]. Having the flexibility to tack quickly to more cost-effective or effective options is a necessity in this space. Through the development and testing completed at the end of Phase II, we were able to make modifications to the application and identify areas of improvements for future iterations.
 +
 
 +
=== 2021-2022 ===
 +
Project Cyclops has just completed testing on the [[wikipedia:Minimum_viable_product|minimum viable product]] (MVP) by Inspectors from the Health Product Compliance and Border Integrity Unit. We have identified and summarized together the strengths and weaknesses of the [[wikipedia:Minimum_viable_product|MVP]]. With the completion of Stream 2 in the Solutions Fund, we have begun exploration into the following key improvements that would be required to bring the product into production:  
 +
 
 +
* Testing alternative options that would increase text extraction accuracy on the claims
 +
* Improving performance by purchasing larger cloud capacity and speed
 +
 
 +
* Focus on enhancing solution to evaluate curved surfaces.
 +
* Expand scope to include processing of multi-lingual labels.
 +
 
 +
We have identified an alternative product, PowerApps, that has demonstrated great capacity to resolve two of the weaknesses noted above. We have begun discussions and intend to look for funding opportunities that would allow us to initiate the development and testing of the solution this summer and bring the [[wikipedia:Minimum_viable_product|MVP]] closer to production.
 +
 
 +
As the roadmap after Stream II has not been laid out, the implementation of Cyclops after the Solutions Fund process into Stream III and/or production needs to be straightened out. Solutions Fund and senior management within ROEB have indicated broad support on the required efforts for bringing this idea to production, and discussions are ongoing on the path forward.
 +
 
 +
Upon success, Cyclops hopes to scale its use to more program inspectors across Health Canada and into the hands of Canadians across the country to harness Cyclops’ super-human like power to verify licenses and receive additional risk information, empowering them to make educated and informed decisions.  
 +
 
 +
=== Repository ===
 +
Source code for the Stream 2 prototype is available [https://github.com/hc-sc/cyclops here]. Code is not available for the operationalization of the project, however a repository to extract and transform required NHP reference information from the [http://webprod.hc-sc.gc.ca/nhpid-bdipsn/search-rechercheReq.do Natural Health Products Ingredient Database] (NHPID) was created and is available [https://github.com/hc-sc-cyclops/NHPID-Data-ETL here]

Latest revision as of 14:37, 27 July 2021

The Technology and Business Innovation (TBI) under the Planning and Operations Directorate (POD) undertook Project Cyclops funded by the Solutions Fund to bring innovation within Health Canada. Although not a one-eyed superhuman strength mythological creature, Project Cyclops aims to develop an online web application that would enable inspectors to use a digital tool/camera to upload images to assess whether a product label is compliant with Health Canada’s regulations.

The scope of the project was limited to focus on natural health products (NHP), in partnership with Inspectors from Health Product Compliance and Border Integrity Unit within ROEB, and financial and technical support from Solutions Fund and IMSD.

The current label review process is tireless, tedious and time-consuming. The process can have more than 10 long manual steps. An inspector can spend hours reviewing multiple pieces of information on a label, such as medicinal and non-medicinal ingredients, dosage, and claims and risk information comparing that to the licence issued by Health Canada. Such inspections are highly repetitive and require very precise focus and attention to details.  

EXPERIMENTATION

The project was separated into two streams:

  • Steam I was a beginning stage into exploring various options and elements that would be necessary for developing a minimal viable product to be built in order for the solution to deliver the provided result. Steam I first goal was to undertake a feasibility study, identify challenges, options analysis and a Proof of Concept (PoC) model.
  • Stream II first focused on the development of machine learning models, following by the development of the solution pipeline that can incorporate the 4 key areas of focus. The solution pipeline includes the privacy design approach to development, rules governed by statutory constraints with inspections enabled by telecommunications, data source integration and automation, and UX design.

In Stream 2, a prototype was developed and tested through a partnership with developers for Statistics Canada that would allow inspectors to review labels on-site, by:

  • Virtualizing optically represented text on labels,
  • Querying data via existing databases,
  • Identifying and highlighting instances of non-compliances,
  • Visually depicting any issues by overlaying information back onto the application.

BENEFITS TO HEALTH CANADA

The Inspectors monitor compliance and conduct enforcement actives to prevent the sale of non-compliant products and protect the health and safety of Canadians. The tool built through this project will enable inspectors to review more product labels and refocus their efforts on other parts of the inspection process and help increase confidence in Health Canada’s national compliance and enforcement process. Product testing by Inspectors in ROEB informed us that inspectors valued this direction as 80% of testers indicated that a fully functioning prototype would increase confidence in the accuracy and the number of inspections.

While currently only being developed for NHPs, this technology has the potential to scale and be repurposed for inspections across other products regulated by Health Canada, such as consumer products, pesticides, drugs etc. TBI has already started engagements with the Tobacco and Vaping Compliance and Enforcement Program for the development of technology to scan labelling of tobacco products. Furthermore, there is an opportunity to develop a similar product for Canadians to verify health information at points of sale, receive any risk notifications (recalls, advisories etc.) and empower them to make informed decisions on the health products they consume.

EXPERIMENTATION PHASE

Working in the experimentation space, we were able to quickly transform the project from just an idea into a fully developed prototype through rapid approval, disbursement of funds and support from Solutions Fund. We also encountered broad support from Inspectors & management within ROEB, despite the pandemic, to make the efforts necessary to test and collect data.

We recognized how invaluable it was to build partnerships with Industry and other Departments to get the necessary expertise in AI and Machine Learning technologies. Solutions Fund through their various initiatives were able to assist our efforts, and connect us to other departments pursuing similar projects, allowing us to leverage the knowledge from previously conducted. As there are various options in the industry in this highly dynamic and constantly evolving subject, it was important for us to continue researching while doing work in existing phases. The project during its lifecycle shifted 3 times in development from vendors such as Lixar, Statistics Canada and Cistel. Having the flexibility to tack quickly to more cost-effective or effective options is a necessity in this space. Through the development and testing completed at the end of Phase II, we were able to make modifications to the application and identify areas of improvements for future iterations.

2021-2022

Project Cyclops has just completed testing on the minimum viable product (MVP) by Inspectors from the Health Product Compliance and Border Integrity Unit. We have identified and summarized together the strengths and weaknesses of the MVP. With the completion of Stream 2 in the Solutions Fund, we have begun exploration into the following key improvements that would be required to bring the product into production:  

  • Testing alternative options that would increase text extraction accuracy on the claims
  • Improving performance by purchasing larger cloud capacity and speed
  • Focus on enhancing solution to evaluate curved surfaces.
  • Expand scope to include processing of multi-lingual labels.

We have identified an alternative product, PowerApps, that has demonstrated great capacity to resolve two of the weaknesses noted above. We have begun discussions and intend to look for funding opportunities that would allow us to initiate the development and testing of the solution this summer and bring the MVP closer to production.

As the roadmap after Stream II has not been laid out, the implementation of Cyclops after the Solutions Fund process into Stream III and/or production needs to be straightened out. Solutions Fund and senior management within ROEB have indicated broad support on the required efforts for bringing this idea to production, and discussions are ongoing on the path forward.

Upon success, Cyclops hopes to scale its use to more program inspectors across Health Canada and into the hands of Canadians across the country to harness Cyclops’ super-human like power to verify licenses and receive additional risk information, empowering them to make educated and informed decisions.  

Repository

Source code for the Stream 2 prototype is available here. Code is not available for the operationalization of the project, however a repository to extract and transform required NHP reference information from the Natural Health Products Ingredient Database (NHPID) was created and is available here