The Key Takeaway
This Minnetronix case study shows how a traditional heuristic approach can shorten time to market while keeping options open for future AI development. To preserve the identity of our client, we’ll refer to this product and its approach to data as Project X.
In the case of Project X, these processes resulted in:
- Saved cost and timeline by choosing an approach that reduced time to commercialization
- Created value by delivering a high-performance prototype device leveraged by the customer to execute their fundraising strategy
- Drove future value by identifying a strategy to achieve their cloud-based data management and infrastructure requirements
The Role of Data Science During Development
Data science is always a critical step in medical device development; however, collecting the data that many algorithms rely on can be a time consuming and costly undertaking. This is especially true in the medical device field, where regulations lead to a much higher cost of input data. Not only this, but regulations surrounding medical device design and manufacturing make altering of designs more difficult, leading to a much higher cost of change. With these difficulties, it’s common for medical device developers to spend time interpreting the data themselves or working with a dedicated data science vendor.
Not every project requires the immediate hardware and data collection investment necessary for machine learning and neural networks approaches that have become the default for many data science shops. In fact, many problems have solutions that require far less development data to get a commercial device to market on-time and on-budget. With the right planning, the costs of collecting the necessary data for more advanced solutions can become less burdensome after device commercialization. Therefore, decisions should consider the total roadmap for the hardware, software, and data collection of your medical device, enabling your ability to craft your next generation of data science algorithm in the future.
To illustrate how Minnetronix approaches cases like this, all one needs to do is look at our work on Project X. This case study shows how Minnetronix’s proven process ensures you take the best approach to medical device commercialization while considering all of your business constraints. Our data scientists are tightly integrated with our hardware and software development teams, ensuring you scope your algorithm roadmap to support early device commercialization, while enabling future product evolutions.
“Our data scientists are tightly integrated with our hardware and software development teams, ensuring you scope the data science algorithm roadmap to support early device commercialization.”
Project X: A 3D Volumetric Imaging Application
This project is a perfect example of how Minnetronix addresses integration and scaling as needed and when needed. During Project X, we implemented a Traditional Heuristic approach to reach near term business goals while planning for machine learning and AI enablement in the future.
Project X served as an imaging tool for anatomical sizing. It was designed to map the size of various anatomical structures over time to aid in clinical determination. Additional goals for the project included:
- Design a device that can measure the size and temperature of an anatomical feature in 3D space
- Provide a cloud-based data management infrastructure in compliance with regulatory requirements
- Achieve project milestones in support of funding goals
- Implement the device in compliance with applicable regulatory standards
- Validate performance through clinical trials
- Commercialize and improve the device
“Our implemented approach was chosen to reach near term business goals while planning for machine learning and AI enablement in the future.”
Selecting Our Approach
Selecting the right data science approach may seem trivial, but it is critically important to mitigate the cost of change. Decisions made early in the process lead to downstream tradeoffs during development. Therefore, a rigorous front-end analysis must consider tradeoffs between data science and other factors like clinical application, ease or difficulty of deployment, regulatory hurdles, and business or market constraints.
Minnetronix Medical has a proven framework to evaluate tradeoffs and lead to an optimal development process. We consider four generalized approaches to data science, measured with six business and capability characteristics, shown in the image below. After strategically analyzing the data science and commercialization tradeoffs for Project X based on our framework, we determined that the Traditional Heuristic approach was optimal to minimize risk while keeping an eye on more sophisticated approaches in the future.
Assessing the Needs for Project X
We chose the proper approach based on six competing variables and tweaked it to fit our desired goal.
Input Data Efficiency
Many algorithms rely on large amounts of data to be effective. However, in this case, large quantities of accurate, clinical ground truth measurements were difficult to obtain, meaning our design had to make do with less data.
Explainability
This capability was given a medium level of importance. A less explainable algorithm would have higher risk in the regulatory pathway; however, the algorithm didn’t need to be explainable from the user’s standpoint.
Computation Speed
High computation speed was of medium importance, given that the calculation would be done frequently on a tablet. The goal was to yield an output within 30 seconds.
Cost and Timeline Certainty
For Project X, the device cost and timeline were crucial. A working prototype device was vital to achieve funding milestones necessary to continue the project towards commercial production.
Performance Ceiling
The device needed to perform better than the current standard. In many instances, outperforming the current standard employed by humans is a high bar to clear. However, in this case, we identified accuracy gaps in the current standard. As a result, the capability ceiling was designated as a medium priority.
Post-Market Growth
Though a working prototype was of the utmost importance right away, new features and improvements post-commercialization were emphasized by the client, and therefore this category was given a higher priority.
Following the Minnetronix Process
The project followed the typical Minnetronix data science process, specifically designed to identify gaps and focus work on creating a viable clinical product.
Our team follows the approach that best addresses your business needs, while keeping in mind the high cost of input data and design changes while developing medical devices. Our approach relies on a trusted seven-step process that enables you to make the best decisions aligned to your specific needs based on factors including data availability, product requirements, business constraints, regulatory strategy, and more. With Minnetronix’s proven process, we can plan for the future and do more with less.
Normally, a static imaging project would be a good application of machine learning, specifically for neural networks. However, the team quickly identified the lack of data which made this difficult. In addition, business constraints led to increased emphasis on cost and timeline. As a result, the traditional heuristic category was chosen. This prevented a costly foray into machine learning that would have deviated from the customer’s primary goal to achieve milestones for commercialization.
Since the model chosen was a traditional heuristic, little data collection was necessary for development. The fixtures were designed with known geometries, and current techniques were applied to determine the current standard for accuracy. The algorithm could then be evaluated with this current accuracy standard.
The data was small enough that detailed organization of the development data was not necessary. However, the customer desired the device to have the ability to interface with Electronic Medical Record (EMR) & Electronic Health Record (EHR) information as part of the workflow when using the scanner and have a cloud-based infrastructure to store scans from the devices and provide physicians an interface to review the data. This also would enable potential development of more complex data science algorithms post-commercialization, and required the development of a workflow and software architecture in compliance with HIPPA and FDA requirements. Minnetronix evaluated options to achieve these goals and defined a path forward to align with the project requirements.
This project had multiple design problems to solve, including accurate 3D imaging, accurate thermal imaging, aligning, and contouring the thermal image onto the 3D scan, and measurement.
Minnetronix explored solutions for each problem–noting pros and cons. Once the individual solutions were identified, the overall solution was evaluated at a system level. Tradeoffs of the combined solution options were then evaluated against the product and customer goals. Ultimately, we selected a stereoscopic vision camera module and a thermal imager module with a customized mechanical housing to achieve the cost, size, timeline, and optical alignment required by the product.
Once the solution space and component selections were identified, we built a prototype device leveraging laboratory optical mounts and 3D printed calibration fixtures with known geometries and thermal markers. This prototype was able to evaluate and optimize the stereoscopic camera and thermal imager configuration; it also was able to develop and implement the calibration process to align the images from the two sensors into a multi-spectral 3D model.
The prototype software was tested against the fixtures and a model representing the type of anatomical features a physician would be using this device to measure. Afterward, the process moved to the implementation and verification phases.
During this phase, Minnetronix began the mechanical modeling of a customized mechanical housing, including tolerancing of the optomechanical stackup of the two imagers in the system. Verification of the custom design is still yet to be completed.
Utilizing a Partnership
The current algorithm performs to the standard required by the customer. The solution created by Minnetronix performs the task to the desired level of accuracy while minimizing the cost and timeline to market.
Project X perfectly demonstrates Minnetronix’s ability to get a product to market while continuing to leave the door open for machine learning in the future. Our partners find this expertise in data science as part of our core services, reducing the need to interpret large amounts of data themselves or seek another partner for additional data science additional data science help.
Minnetronix provides unparalleled value by utilizing our data science selection and development processes. In the case of Project X, we were able to work early in the development of the medical device and choose an approach that not only takes into account peak performance of an algorithm, but also weighs business, data, and regulatory constraints.
The development process was also invaluable. Our proven seven-step process accomplished our goals for the algorithm, without adding unnecessary risks and complexities inherent to more novel techniques.
In the case of Project X, these processes resulted in:
- Saved cost and timeline by choosing an approach that reduced time to commercialization.
- Created value by delivering a high-performance prototype device leveraged by the customer to execute their fundraising strategy
- Drove future value by identifying a strategy to achieve their cloud-based data management and infrastructure requirements
We know how to optimize inputs to minimize hardware requirements and achieve performance. When you work with Minnetronix, you’ll know that we choose the best data science path to maximize the data you have.
Interested in learning more? Click here to contact us!