Medical AI, ML & Data Science Services
Building regulatory compliant medical AI
Camgenium's team of experts have been building regulatory compliant AI solutions for over ten years. Our team of data science engineers not only build sophisticated AI & ML solutions but also have the regulatory knowledge to ensure every model is built to regulatory standards. Camgenium's advanced imaging skills, data science and engineering backgrounds coupled with deep medical knowledge, have enabled us to build world-class AI/ML products.
Already have an AI solution? We can take your existing model and wrap it in a compliant package for deployment in a healthcare setting. We can then host it on your behalf on our regulatory compliant systems.
Over 10 years' experience in medical AI & ML development
Camgenium started its AI journey in 2015. Our first product was based on a slow research prototype that used the Caffe framework to analyse ultrasound images. We re-engineered this prototype to create a medical device that was more accurate and analysed real-time video images. The original algorithm had been developed using a training set of over 100,000 ultrasound images labelled by Sonographers and these images were used to validate the product.
The software automatically identified foetal anatomy and removed video artefacts such as personal identifiable data. It was used as a training aid for new Sonographers to help them learn how to capture the required images needed as part of routine pregnancy scans. The video below demonstrates the prototype developed over ten years ago.
My role in the Data Science team involves the processing of datasets to gain valuable insights, and to develop models to aid clinicians and hospitals in healthcare environments.
Anna Vuolo
Head of Medical AI, Camgenium
Intelligent medical devices
AI will enhance the capability and flexibility of medical devices in the future, but there are significant engineering challenges. When the Internet of Things (IoT) was first named in 1999 by Kevin Ashton, the vision of the network was billions of sensors connected to powerful central computers that would perform signal analysis and computation. Today, we aim to design devices that can send actionable information rather than raw data for safety, to minimise network traffic and the energy used by each IoT device (transmitting data often requires much more energy than local analysis).
Next generation networks will distribute the intelligence to their edge, in other words to the devices. This is possible because of the new capabilities offered by next generation multiprotocol chips, such as those produced by Nordic Semiconductor.
It is now possible to run ML tasks in real time at low power in the device, allowing information to be derived from data at source. Multi-core chipsets are transforming what is possible. However, designing co-processing applications that use multiple cores is complex and this challenge is made even more difficult by having to work to medical standards.
Camgenium’s team of engineering experts have the knowledge and skills to use these sophisticated new technologies. The team has years of experience designing advanced world leading embedded platform products. We have built a skilled team of Cambridge University AI/ML engineers working with huge data to inform models that create next generation medical devices. Our team understand what makes a successful medical AI model and know the appropriate steps that need to be taken to ensure the data is successfully curated and is ready to be modelled. Camgenium's expert team is skilled in collecting data from multiple sources and varied systems.
Embedding medical AI into products
At Camgenium, we have extensive experience of developing wide-ranging algorithms from massive data sets (well over 100 million patient records from multiple countries around the world). We can assist with development patterns to conform to medical device standards and advise on appropriate validation experiments to enable submissions to be made for regulatory certification of the algorithm.
Once the algorithm has been developed, the code often needs to be wrapped in an application that provides a way for users to interact with it so they may benefit from the insights it provides. Wrappers must be regulatory compliant, particularly in the way in which they handle personal information. Users can include both individual people and computer systems. The Camgenium platform has successfully been used to wrap algorithms to make them accessible and usable by clinicians within healthcare settings and to interface them to fortressed healthcare IT systems.
Camgenium not only has the data science expertise in house, but we also have medical experts who understand the regulatory and ethical requirements of developing medical AI. As the Camgenium platform is so extensive, minimal work is required to embed algorithms into existing devices and turn them into usable products. This means that the time to market can be minimised to allow you to take full benefit of any first mover advantage.
Providing secure and compliant access to real-time data
To deploy AI algorithms, access to data must be negotiated. Obtaining the required information governance approvals can prove very difficult and certification to rigorous standards is normally required.
Camgenium is trusted with sensitive healthcare data worldwide. We work to mandated standards, including contracted standards in the UK which are significantly higher than ISO 27001 and allow us to access and process many millions of healthcare records.
We run third party AI algorithms in our IT estate to enable them to become part of standard healthcare processes with the access they need to current healthcare data.
Testing medical AI
Testing medical AI models to ensure accuracy and validity is paramount. Camgenium’s team rigorously test each medical AI model, employing industry best practice such as testing on a separate dataset to the training data to understand how well the model performs on unseen data, prior to deployment in a live environment.
The examples below demonstrate the importance of testing medical AI to ensure accuracy. Both models are trying to predict the presence of a disease, which exists in 50 out of 1000 people.
Model 1 hasn't predicted any cases of the disease (0 in the predicted diseased boxes) and hasn't correctly identified any of the 50 cases of disease. It has a high accuracy score of 0.95. Model 2 on the other hand makes some correct and incorrect predictions, giving a lower accuracy score of 0.94.
On first look, model 1 might seem better, but a higher ROC score of 0.8 in model 2 is the more important metric to examine.
Developing bespoke sophisticated medical algorithms
Camgenium provides algorithm development services. We have an expert team of Cambridge University engineers who have trained in machine learning and artificial intelligence. This team has developed award winning products with exciting algorithms at their hearts. We have a great deal of expertise in processing medical data, and we understand many of the challenges of working with standard medical data sets. We also have significant expertise in visualisation of results.
Camgenium develops pioneering medical algorithms for clients but also has its own platform. Camgenium's AI-powered medical analytics tools assess hospital performance based on risk-adjusted clinical outcomes. The technology has been in use in hospital settings worldwide for over 17 years. In that time, over 500 million patient records have been analysed to create next-generation AI models for the NHS and hospital settings globally. These tools have enabled Camgenium to have access to large NHS datasets (subject to necessary provisions and NHS approval).
Develop your bespoke AI or we can integrate your existing algorithm into your medical device product.