Interest in AI has been increasing in recent years - with healthcare at the pioneering forefront. The field of cardiovascular imaging is undergoing a paradigm shift toward AI-driven platforms. These diverse algorithms can seamlessly analyze information and automate manual tasks to help improve patient outcomes with better diagnoses, save lives, and solve limitations.
Numerous commercial solutions using AI are now available. However, similarly to human knowledge, not all AI systems are created equally, and not all have the same abilities. AI platforms should meet good AI practice, and have the ability to solve problems, learn, and be future proof. They should be ethical, robust, and show evidence of rigorous testing. Additionally, all AI platforms should be easy to implement and meet regulatory and international standards.
Keep reading to learn more about key features to consider when investing in an AI system for echocardiography.
Be Beneficial and Enhance Diagnostic Capabilities
Reducing hospitalizations and minimizing diagnostic misses remains a high priority. AI systems can help provide quality care and improve outcomes in alignment with the Quadruple Aim.
Within the realm of echocardiography, AI can be adopted in clinical workflow to automate the assessment of cardiac function, which would usually take time and expertise. The result is a revolutionary system that can support faster heart disease detection, minimize variability caused from different levels of clinicians’ experience, and enhance diagnostic capabilities.
But it is not just automation that makes a standout system. Using Ultromics’ AI service, cardiologists have managed to predict outcomes, including heart failure. The AI system leverages machine learning and deep learning to help clinicians make patient predictions and prevent adverse outcomes, using cutting-edge insights and multi-dimensional data trained on 10-years of outcome data.
In a study published in JASE, cardiologists at MedStar and the University of Chicago looked at 870 patients at 13 medical centres in four world regions.  Ultromics was able to predict heart failure when manual analysis could not. It could also eliminate variability, which affects diagnostic accuracy. Preliminary findings with Mayo Clinic, presented at ASE 2021, also found cardiologists who used Ultromics could detect HFpEF with 96% accuracy. In addition, a study published in JACC Imaging, reported cardiologists could detect CAD with 10% more sensitivity.
These results provide a new benchmark for diagnostic accuracy in detecting heart disease and heart failure in echocardiography.
Avoid Unfair Bias and Be Ethical
The quantity and quality of data used to train the AI system has a significant impact on the accuracy and reasonableness of the AI outputs. For AI to have optimal patient benefit, it needs to undergo rigorous design with social and ethical considerations at every stage of its development.
That said, the AI may only be as good as the data used to develop it. AI systems must consider sources of bias in echocardiographic data, which can undermine performance and limit applicability to wider populations.
Important factors such as geographical and ethnic heterogeneity in data collection, spread of disease types and severity, image quality, range of heart rates, sampling frequencies, and ultrasound vendor inclusion should all be considered.
At Ultromics, our AI algorithms are developed using several key principles to minimize the risk of bias and maximize generalizability. This includes obtaining data from international, multi-site datasets to maximise diversity, using real-world information obtained from routine clinical practice, and collecting data from challenging imaging scenarios, including stress and limited echocardiograms. Additionally, we ensure
a balance of ultrasound machine vendors and models to account for heterogeneity in image acquisition parameters.
The team behind Ultromics includes world-renowned scientists and engineers who specialize in pioneering deep learning, machine learning, medicine, clinical research, medical device development, and regulations across the US, UK, and Europe. The diversity in expertise of our staff not only ensures technical competence, but also works to facilitate collaborative efforts when addressing potential bias in AI algorithms.
Meet Compliance and International Standards
One of the most important considerations when onboarding any new medical technology is regulatory approval. Clinics need to confirm that the devices have been cleared not only by medical associations, such as the Food and Drug Administration (FDA), but also meet information security standards.
Ultromics’ AI solutions have been specifically designed to meet the requirements set forth by international regulators and governing bodies. This includes, but is not limited to, the principles outlined jointly by the Food and Drug agency and Medicines and Healthcare Products Agency in “Good Machine Learning Practice for Medical Device Development.”
Ultromics is part of the American Society of Echocardiography (ASE) AI task force which sets good AI principles in the realm of echocardiography.
Additionally, all of our products and technical infrastructure are compliant with international standards on information security (ISO27001), software development (IEC/ISO62304) risk management (ISO14971) and usability (ISO62366-1).
Ultromics is also HIPAA compliant, with rigorous VPN-based security systems in place to protect patient information. Whereas on-premise devices might have variable levels of security, cloud-based technology allows for protection through VPN, which can ensure that all patient data stays safe.
Have a Rigorous Design and be Built for Accuracy
AI providers must undergo rigorous design and testing of their algorithms and datasets to underpin a robust and valid AI system. Ultromics’ system was built by world-leading experts on machine learning from the University of Oxford (UK) on 10 years of outcomes data, and has undergone a rigorous systematic review of the accuracy of its AI for the diagnosis of cardiovascular disease at every stage of the AI pipeline.
The auto-contouring models have been developed to automatically delineate and contour the endocardial border of the LV in echocardiographic images in the A2C, A3C, and A4C views, passed through the 2D U-Net auto-contouring model, which then learns abstract representations of the given input. Performance of the auto-contouring models demonstrated acceptable performance with mean Dice scores across all images of 0.892, 0.833 and 0.921 for A2C, A3C and A4C views, respectively.
When evaluating Ultromics’ AI platform, diagnostic accuracy was validated primarily by comparing model predictions to competitor leading software, using hundreds of thousands of images across international, multi-site datasets from a range of key data sources capture the different care settings along the pathway, including different conditions such as primary care, social care and cardio-oncology, and clinically indicated routine echocardiograms.
The AI outputs in EchoGo have been shown to more accurately correlate with patient outcomes than manual reporting. One study showed that fully automated measurements made by EchoGo were more reliable than human interpretation. 
Have Robust Validation and Real-World Testing
Validating AI platforms through independent, real world testing is another essential aspect to consider when investing. This is separate to the data used to train the algorithm, and ensures not only accuracy, but device usability across the population.
At Ultromics, we validate our technology through large scale external validation datasets across many healthcare organizations around the globe. Medical devices are tested on datasets unseen by the AI with demographics and characteristics that are distinct from training datasets. Datasets are also measured to test clinical performance and assess the safety, efficacy, and potential impact of the AI platform when integrated alongside standards of medical practice.
Our validation has been acquired from hospitals with a range of sizes, types of operators and ultrasound vendor equipment representative of “real world” echocardiography, with over 50 sites, including the NHS UK (ClinicalTrials.gov identifier: NCT03674255), MedStar (The international MedStar World Alliance Society of Echocardiography II COVID study), and over 50k images at the Oregon Health Sciences University (OHSU).
This robust validation has allowed us to develop trusted and highly accurate models which have superior outcome prediction, and which are relevant and future proof, leading to better decisions and greater confidence in our accuracy.
Uphold AI Standards, Performance, and Scientific Excellence
Technological innovation is rooted in the scientific method and a commitment to open inquiry, intellectual rigor, integrity, and collaboration. AI tools have the potential to unlock new realms of scientific research and knowledge in many different clinical applications. Ultromics aspires to high standards of scientific excellence as we work to progress AI development.
After designing and validating AI solutions, post market surveillance within real-world clinical practice is essential to ensuring continued performance. This helps to identify re-training needs, evaluate equity of outcomes across the population, and maintain safety.
This is why Ultromics aims to conduct as many trials as possible. We continue to be involved in trials seeking to obtain the best assessment of performance, under the highest standards, to assess equity of outcomes and patient safety.
We are also collaborating with leading experts to retrospectively assess performance. Our goals are to continually evaluate device safety and performance across a range of diseases and conditions.
Another consideration is whether the AI is continuously innovating and evolving. With accuracy, rigorous testing, and security concerns all being prominent factors while researching AI diagnostic technology, it may be easy to overlook the importance of innovation. However, when onboarding any new technology, ensuring it stays future-proof is key for long-term return on investment.
The AI in EchoGo is constantly evolving by collecting diverse data across geographic regions and challenging imaging scenarios. This cloud-based technology is always learning, which improves accuracy and usability
Be Easy to Implement into Clinical Workflow
Beyond the foundations of clinical accuracy, performance, and compliance, AI-based medical devices should be easy for practitioners to use. To avoid unnecessary and potentially expensive training, there should be a small learning curve.
Additionally, the technology itself should be straightforward to onboard and cause no disruption to existing clinical workflow.
At Ultromics, we have a streamlined onboarding process that ensures ease of implementation and use for all practitioners. It also allows for a wider variety of clinicians to perform and analyze routine reporting, from nurses to specialists.
Adopting responsible echo AI is a goal for any practice looking to improve patient outcomes, which is why it’s important to complete thorough research before choosing the best option. Top considerations, such as its accuracy, design, validation, security, and regulatory compliance, are all key to ensuring that your practice is onboarding the right technology. Ease of imaging is also important, as any new technology should be an improvement on previous solutions. The AI must also be future proof in order to achieve a return on investment and avoid needing to onboard new technologies more frequently.
These are the top goals that Ultromics continues to not only focus on, but also achieve. To uncover more about how EchoGo’s responsible AI is revolutionizing echocardiography, download our free Heart Failure Guide today.