AI tools for medical device development
Published: August 2024
Grant Howarth
Interaction Design Consultant
Peter Fischbacher
Design Consultant
In recent years AI tools have become increasingly powerful, with new use cases appearing almost weekly. They’re now well integrated into our daily lives from smart home devices to social media posts tailored for us, and it’s fuelling a growing appetite to incorporate further tools into our design processes.
As designers, our goal is to deliver projects on time while creating medical devices that are effective, appealing, usable, and safe. To continue to achieve this, we are continually evaluating new technologies and trends.
However, the use of AI raises questions: What should be considered when using AI tools to achieve class-leading medical design? What are the ethical considerations and risks to intellectual property?
In this article, we share what is important to know when evaluating AI tools and their applications in medical device development.
Getting started with AI
Keeping up with the pace of change in AI
The pace of AI development is extremely fast. If you’ve previously tried AI tools and found them lacking, consider giving them another chance. It’s worthwhile to set a reminder for re-evaluation every 3-6 months to stay current with their capabilities. At Shore, we’re continually assessing tools and their practical applications as part of our AI strategy.
Spending time to learn how to use AI tools
Your initial experience with current-generation AI tools might be disappointing. The images and text they produce may not align with your intentions or prove useful. Like any tool, these require practice and learning to understand how to input data for your desired output. However, as AI tools and the means of interacting with them improve, they are likely to get better at understanding your requests and provide you with more control over the outcome.
Do AI tools live up to the hype?
Look through social media or any design news site and they’re flooded with AI tools making big claims about their capabilities. Often, these claims are exaggerated or show the tool working well in a very specific area.
In the design world, these claims often miss the fact that design involves more than creating attractive-looking concept renders. Medical device development is a complex and varied process, requiring a broad range of expertise: from engineering and industrial design to regulatory and human factors considerations.
While AI tools can assist with some parts of this process, they are currently unable to harmonise all these considerations effectively.
What to consider when adopting AI
Confidentiality and Intellectual Property
A potential concern with many AI tools lies in how they handle data submitted via prompts. Some tools use this data to train future models, which poses a risk of public accessibility down the line. Certain image generation tools may also display user-created images on their websites. This becomes particularly worrisome when a sketch or reference image is involved, as such information could inadvertently end up published.
In short: don’t put in what you wouldn’t want others to see.
Most tools offer a premium subscription tier with enhanced privacy features, but it’s crucial to carefully review the terms and conditions of use. Notably, if sketches or concepts are discovered to have been publicly shared online, this could potentially invalidate future patents.
Data submitted to large language models (LLMs), like ChatGPT, when processing results from human factors testing must be handled with extreme care to avoid breaching patient confidentiality. In some cases, the benefits may be outweighed by the risks. Regulations including the General Data Protection Regulation (GDPR), California Consumer Privacy Act (CCPA) and the Health Insurance Portability and Accountability Act (HIPAA) must be strictly followed when handling any Protected Health Information (PHI).
Reliability of the output
AI tools are typically trained on a vast amount of data from various sources, which means they can inherit biases present in the training data.
The training data also impacts the AI’s ability to suggest new ideas. These tools are more likely to propose solutions or ideas that already exist. It’s important to ensure that any output, whether text or images, does not directly copy elements from other sources.
Current-generation LLMs can suffer from ‘hallucinations’, where the model confidently states information that is not true. Often this is easy to detect, for instance, a product or service may be suggested that doesn’t exist. However, sometimes this might not be so obvious, particularly if the subject area is one that you are not familiar with. For this reason, these models must never be used in situations where it is not possible to verify their output.
Similarly, current AI image generation tools are good at producing things that look convincing but may be technically unfeasible, non-sensical or unmanufacturable. Image generation tools are trained on data that will include products that are poorly designed. Care needs to be taken to avoid copying bad design features that are suggested by these tools.
Current AI use cases
Despite the aforementioned hype and potential pitfalls, AI tools are already helping us be more productive day to day. Primarily they help us by improving efficiency, and by supporting creativity.
Efficiency
AI tools can increase efficiency in several areas, saving time and effort. Some examples we have tried include:
- Using LLMs when researching new topics, understanding concepts or summarising information – for example, recent literature on specific topics or current industry best practices.
- Improved audio transcription via ChatGPT or other AI tools.
- Using image tools to rapidly iterate sketched ideas (e.g. turning sketches into illustrations or product renders), and at any point creating variants or adaptations based on prompts.
- Creating materials or backdrops for product visualisation and renders allows you to visualise the product in various contexts.
- Image processing including upscaling, expanding and removing items from images.
A pen-style medical device created using the AI tool, Midjourney. The animation showcases the evolution from initial results to more refined concept renders achieved by tweaking prompts and other settings.
Creativity
AI tools also have the potential to aid the creative process for example:
- Overcoming the ‘blank canvas’ barrier.
- Supporting brainstorming and ideation due to its ability to rapidly create a multitude of stimulating images, feasible or not, that can spark ideas.
- Creating mood boards from text and image prompts.
AI tools offer exciting possibilities for improving the medical device design process. Still, it’s important to remember that they can’t compare to a team’s cumulative hundreds of years of experience in medical design. The integration of AI tools into our workflows should complement, not replace, the deep expertise and human element required.
It’s also important to recognise that we are in the early days of AI, where the landscape is fragmented by niche products tackling well-defined or specific problems. While it’s possible to develop workflows as individual tools, industry leaders will eventually integrate broader functionalities.
