Dr Carolyn Lam
Senior Consultant Cardiologist, Principal Lead of Clinical Trials & Trial Networks, National Heart Centre of Singapore
 

Setting sight on breakthrough diagnostics

KEY TAKEAWAYS

  • AI has enabled the generation of fully automated, fast and reproducible echocardiographic analysis supporting current traditional diagnostic tools for heart failure
  • The combination of cardiac biomarkers NT-proBNP & BNP) and AI-enabled ECHO interpretation is the ideal diagnostic tool for heart failure
  • Further advancements in medical AI enable deep learning models to be developed for greater diagnostic/ predictive precision

Professor Carolyn Lam, Senior Consultant Cardiologist and Principal Lead of Clinical Trials & Trial Networks at the National Heart Centre of Singapore, shares how artificial intelligence (AI) can enhance future diagnostic strategies by aiding current traditional diagnostic tools for heart failure, delivering accurate and timely diagnosis.

How large and dark is the “cloud” of heart failure?

Heart failure affects over 64 million people worldwide. The burden of disease is tremendous, and heart failure is the top cause of hospitalizations among the elderly, with an estimated cost of USD $108 billion per year. Delayed diagnosis and under-treatment contribute to the high hospitalization burden of heart failure worldwide.

The diagnosis of heart failure can be challenging, as patients with this disease experience non-specific symptoms and signs such as shortness of breath on exertion, weakness, and oedema. It is estimated that 1 in 6 people greater than 65 years old presenting to primary care with breathlessness will have unrecognised heart failure.

What is the “Silver Lining” in the diagnosis of heart failure?

Due to heart failure symptoms often being non-specific, current guidelines recommend the measurement of Natriuretic Peptide cardiac biomarkers (NTproBNP/ BNP) in combination with cardiac imaging of echocardiography for the initial diagnosis of heart failure, which provide objective evidence of cardiac dysfunction.

The aim to perfect the combination of natriuretic peptide and echocardiography diagnostic tools is important. Often biomarkers may be influenced by comorbid conditions, such as atrial fibrillation, advanced age, renal failure and obesity making results difficult to interpret. Echocardiography too has its inherent limitations as it is difficult to analyze, requiring highly trained specialists to interpret the images in a very manual, time-consuming and error-prone process.

Advancements in AI enable the generation of fully automated, fast and reproducible echocardiographic analysis. Madani et al (2018) showed that deep learning models could enable the rapid and accurate classification of echocardiographic views for machine recognition of the chambers of the heart from different angles. Zhang et al (2018) extended this in a proof-of-concept pipeline for echocardiographic interpretation, from view identification to image segmentation, chamber quantification and disease detection. Ouyang et al (2020) then went on to show that beat-to-beat predictions of left ventricular ejection fraction could be achieved, with a variance comparable to that of human readers. AI software has been developed, turning a highly manual process that involves 30 minutes per study and requires 250 clicks by the sonographer, to a fully automated process taking only two minutes with just one click of a finger (http://us2.ai/).

“Calming the Storm”- What is the ideal diagnostic combination?

The combination of circulating Natriuretic Peptide cardiac biomarkers (NT-proBNP/ BNP) and AI-enabled echocardiographic interpretation is the ideal diagnostic tool for heart failure.

In the past, combined interpretation of these two tests was not possible since they were performed independent of each other. Currently, many hospitals have linked electronic health records and picture archiving and communication systems (PACS) which allow combined interpretation of both tests.

The synergistic use of AI and the clinical decision-making skills of physicians will enable efficient and timely diagnosis of heart failure. This will lead to reduced hospitalisation and hospital costs, reduced waiting lists, and efficient use of hospital staff and resources, a paradigm shift in the management of heart failure!

What are the possible impediments of AI?

The use of AI can drastically change the landscape of heart failure diagnosis. To maximise the benefits of AI, certain factors need to be considered.

Regulatory testing is needed to minimise the risk of errors in the machine algorithms. Huge datasets are needed in the creation of algorithms to account for biological diversity. Equitable access, data privacy and data security are also important factors to be reviewed to ensure the success of this technology.

Future has a “Rainbow of opportunities”

AI has enabled the generation of fully automated, fast and reproducible echocardiographic analysis supporting current traditional diagnostic tools for heart failure. The combination of cardiac biomarkers Natriuretic Peptides and AI-enabled echocardiographic interpretation is the ideal diagnostic tool for heart failure, working harmoniously along with human clinical decision-making expertise, which will transform the diagnosis of heart failure (Image 1).

Image 1: Combining echo + circulating biomarkers with AI for the ideal HF diagnostic panel

Further advancements in medical AI enable deep learning models to be developed for greater diagnostic/ predictive precision than ever achieved before, with recognition of patterns of disease beyond the human eye. The availability of point-of-care testing for both NT-proBNP and echocardiography (using mobile echo probes connected to handheld smart devices) can foreseeably bring these novel AI-enabled tools to the primary care setting. It is even be possible to use AI to guide an untrained person to acquire the images for echocardiography and to generate a fully automated and fully annotated report within minutes (Image 2).

Image 2: AI-assisted echocardiography. A green light indicates that the probe was placed in the correct position, allowing view recognition and assessment of the machine.


References

Ali Madani, Ramy Arnaout, Mohammad Mofrad and Rima Arnaout: npj Digital Medicine (2018) 1:6 ; doi:10.1038/s41746-017-0013-1

Ouyang D, He B, Ghrobani A, Yuan N, Ebinger J, Langlotz CP, et al 2020 Apr;580(7802):252-256. doi: 10.1038/s41586-020-2145-8.  Epub 2020 Mar 25

Updates from Leading Experts

STRONG-HF: Post-discharge Heart Failure Management and Implementation of GDMT Heart Failure Therapy

Prof Alexandre Mebazaa shares the key results and highlights from the STRONG-HF study that assessed the safety, tolerability of optimization of guideline directed medical therapy. [...]

Read...

CANVAS study: NT-proBNP and CVD risk reduction

Can you give us a short description of the CANVAS program? The CANVAS program was a set of two trials, whose goal was to evaluate the role of an SGLT2...

1 July 2021
Dr James L Januzzi