Data Science Case Studies
AI Neural Network (Deep Learning)
When wielded under the right conditions, deep learning can be a powerful tool to produce predictions that meet or even exceed human expert level performance. These methods are very reliant on large amounts of standardized training data.
In this case study of a CT segmentation project, we show how careful planning and decision-making during algorithm design minimized the costly task of data collection, reduced the risk of timeline creep, and produced a high-performance deep learning model.
Traditional Machine Learning
Not all machine learning algorithms are equal, both in complexity and requirements. Oftentimes, traditional machine learning models (KNN, SVM, Bayesian Networks, etc) have lower data and processing power requirements than deep learning models but can achieve similar performance. As a result, they can be used as a mechanism for evaluation of data and project feasibility before investing more time and resources in a deep learning model.
In this case study of a vital sign monitoring project, we show how models were leveraged as an evaluation tool for key data feature detection and project feasibility. The findings were ultimately used to deeply understand device requirements before design work had even begun.
Sometimes the right tool for the job isn’t a machine learning algorithm at all. When data is scarce, and timelines are strict, traditional heuristics can provide good solutions to get a product to market. After market adoption, the data necessary for machine learning models becomes more cost-effective. However, it’s important to begin planning for this early, since the data you collect now will determine what you can accomplish with machine learning in the future.
In this case study of an anatomical imaging project, traditional algorithms were used to get a product to market while planning for machine learning in the future. This enabled our customer to solidify the scope of the required machine learning development and cover the costs of that development.
The utility of data isn’t limited to building algorithms. With the right skills applied, data can be leveraged to support clinical efficacy of a device, identify, and solve development issues, and inform future device design. Data scientists can bring a different perspective to multifactorial development issues that are difficult to isolate. In addition, data scientists have the statistical background necessary to interface with clinical data to extract meaningful insights.
In this case study of a fluids-based medical device, we show how data scientists were leveraged across the project lifecycle, from clinical trials to product development. These data scientists worked within the design team to move gating items for the project and ensure the value and efficacy of the device was easily visible to customers and clinicians.
Choosing The Right Approach Tailored To Your Needs
Minnetronix leverages a trusted framework to select an approach to data science that minimizes future design and business risk. We consider four general approaches to data science. Each approach provides benefits and drawbacks, according to six competing characteristics.
The Minnetronix Way
For 25 years, Minnetronix has been helping companies design and manufacture medical devices. We choose the right approach to data science problems to prevent costly redesign in the future and form the best path to the business result you need. Minnetronix executes a proven process to get data science solutions integrated with devices in a timely and cost-effective manner.
Your Roadmap to Success
Regardless of which approach to data science is best, every project follows Minnetronix’s trusted seven-step process. This process produces valuable insights and viable products for our clients.
Contrary to many design firms, we can control the data produced by the device itself. This offers an avenue for cost savings and design optimization that is impossible when data science and medical device development are siloed. Involving data scientists as early as possible in the development cycle is important, since the data you collect now will determine what you can accomplish in the future.
Get our Free Whitepaper, Demystifying Data Science in Medical Devices