Challenge

The client is a global life sciences company providing clinical research services for GI-related ailments. With their primary services including clinical trial delivery and medical imaging solutions, the client used manual processing for clinical imaging and annotation management, which proved to be time consuming, incongruous and expensive. Our client set a goal to provide the most accurate, standardized, and timely results, the client sought to modernize their existing data augmentation process using machine-learning (ML) based models.

Prior to engaging with Onebridge, the client established a plan to harness the power of machine learning models to modernize their process of analyzing and classifying biopsy images with a goal of increasing efficiency and reducing cost. The client needed support assessing the quality and readiness of the models before deployment and engaged Onebridge as a development partner to accelerate the project and provide resources to their internal teams to aid the implementation of the new software.

Solution

Phase 1: Business Understanding & Problem Definition

Before the team could kick off the technical aspects of this project, they needed to gain perspective on the current state of development to understand the initial problem and desired outcome.  For this initial phase, the team met with the client’s respective project owner and subject matter expert (SME,) to understand their ideal solution.  This included, familiarizing themselves with supporting software, understanding how the project will be conducted, establishing KPI’s and a definition of success, and completing a comprehensive analysis to attain, understand, clean, and merge the organization’s data.

The model utilizes many parts: GitHub, terraform for deployment automation, Docker for containerization, PostgreSQL database, GIS technologies for image rendering, Python tools for image processing, React.js and typescript for the UI, and a suite of Scala technologies for backend services.

With this extensive list of data tools being utilized, and their lack of in-house data scientists, the client brought in the Onebridge team for their cumulative knowledge of machine learning, data science and code to ensure the model’s readiness for deployment and support their organization in adapting to this change.

‍

Phase 2: Model Design & Development

Now that the team had a full understanding of what this partnership was meant to accomplish, they began the Model & Development phase, where the team established their approach for this project. Using their findings from phase 1, the team defined how they will configure the ideal model for the client. The team considers types and quantities of models to compile their list of candidate models. During this phase they also decide what tools to use and how to allocate the development of the model. For each of the candidate models, the team defines the hyperparameters and outputs so the client can fully understand their options.

‍

Phase 3: Model Output Assessment

In this phase, the model is evaluated according to the KPI’s defined in the initial phase. Each candidate undergoes an in-depth model output assessment where the code and training processing are assessed based on the stated KPI’s.  The results are compiled into a table for the team to compare candidates and discuss the advantages of each choice. The team presents their findings to the client where the partnership discusses and compares candidates, fine tuning details and narrowing down options until the decision makers are confident that the solution is going to satisfy the organization’s needs.

‍

Phase 4: Model Deployment

Once the client has finalized the design of the model, the partnership can begin the Model Deployment phase where the architecture designed to support the ML based model is produced. Because the supporting hardware is typically created prior to the ML model and this task is not typically performed by a team of data scientists, Onebridge supported the client through this phase by deploying the models using containers.  The team chose this software for deployment to ensure the client could access the models when needed and have the option to migrate the model between hosts, even after the engagement has finished.

‍

Phase 5: Monitoring & Maintenance

Because Onebridge aims to support our clients through the entire lifecycle of change, our last project phase is often training. We consider this phase crucial to any partnership to ensure our clients can effectively understand and apply changes even after our partnerships have ended.  To provide our clients with the confidence to navigate new tools and strategies, Onebridge conducts workshops or classroom style trainings and provides users with extensive resources to understand the model.

Outcome

Because the deliverable for this project was knowledge, the result of this engagement was a 90-page report detailing the full assessment of the client’s machine-learning learning model with recommendations for successful implementation going forward.  Onebridge provided the client with the information to understand their product gaps, establishing a foundation to migrate their processing toward complete data intelligence.

As a result of this project, the customer gained new confidence in making informed decisions that allow them to generate more efficient, consistent, and accurate results. The feedback from the client has been consistently positive, with the greatest praise highlighting the structure and thought leadership our partnership has promoted throughout their organization.

To learn more about Onebridge can support your organization, Contact Us here.