At the heart of artificial intelligence: the future might well be based on synthetic cells  Free

Commentary on ‘Reconfigurable perovskite nickelate electronics for artificial intelligence’ by Zhang H. et al., Science 2022;https://doi.org/10.1126/science.abj7943

Technology pervades our everyday lives, and there is no doubt that the future of healthcare is intimately intertwined with high-tech advancements.¹ Skilled medical professionals have the ability to acquire, integrate, and interpret complex information in order to improve individual health, illustrating the human brain power. For many years, researchers therefore have tried to create technology with capabilities of the human brain, to support the medical practitioner in their task and potentially surpass the human in speed and ability to handle the overload of information.^1,2 While the first artificial neural networks were proposed in the 1940s,³ it is only in recent years that interest of the information technology industry has led to the development of what is referred to as Deep Learning (DL): complex algorithms mimicking neuronal activity and with the ability to learn by example from large datasets.⁴ These tools require large (super-) computing infrastructures to be trained, given that our current computing architectures far from resemble the brain anatomy. To solve this technology mismatch, an intelligent material that learns by physically altering itself, similar to how the human brain develops and learns, could provide a basis of a novel generation of computers.⁴ In contrast, current computing technology only simulate this behaviour given that it has no intrinsic capacity to learn and modify from experience or grow and shrink when required.⁴ To address this fundamental gap of task and tool, in an innovative study, Zhang et al.⁵ present an adaptable new device that can transform into all the key electric components needed for artificial intelligence (AI) hardware, for potential use in robotics and autonomous systems. They demonstrated the on-demand creation of artificial neurons, synapses, and memory capacitors in post-fabricated perovskite NdNiO₃ devices that can be simply reconfigured for a specific purpose by single-shot electric pulses.⁵ The authors believe that this adaptability would allow the device to take on all of the functions that are necessary to build a true brain-inspired computer, not only simulating neuronal activity, as is the case with current computers, but truly providing both the anatomical and functional brain capabilities. This opens up avenues for AI technology that continuously learns, grows as needed, and gradually improves, something that current AI systems simply cannot do. It is an important step towards more open-ended, more capable AI.

Fuelled by these advances, in the near future, AI is expected to help realize the promise of precision medicine in three major areas: (i) disease prevention, (ii) personalized diagnosis, and (iii) personalized treatment. In all these three areas, AI and machine learning (ML) provide algorithms that are increasingly valuable to help physicians in their everyday work, particularly in cardiovascular medicine² (Figure 1).

Figure 1

Artificial intelligence (AI) in cardiovascular medicine: schematic outline representing the information flow and inter-links between different data sources. Currently, there are a multitude of different structured and unstructured Big Data sources in cardiovascular medicine, including ‘omics’ data (genomics, proteomics, etc.), high resolution medical imaging, and electrocardiogram data, biosensors, wearables, continuous physiologic metrics, and electronic health records. This Big Data holds huge potential for the use of sophisticated analysis by AI and machine learning in cardiology, improving clinical diagnosis, risk prediction, therapy selection, health systems workflow, and primary and secondary cardiovascular health prevention. Thus, the incorporation of AI into cardiovascular medicine will affect all aspects of this field, from research and development to clinical practice to population health. This figure was created with BioRender.com.

Open in new tabDownload slide

In cardiology, AI, and specifically ML, is already becoming clinically established in cardiac image analysis due the fact that the DL has proven to be highly efficient at extracting spatial and temporal associations from large databases.⁶ Additionally, combining ML with biophysical models of the heart, enables the integration of pre-existing knowledge of human anatomy and physiology.^6,7 AI has been applied to all cardiac imaging modalities, from 2D and 3D images to temporal sequences derived from MR, CT, nuclear imaging, ultrasonography, or electrocardiographic imaging.⁶ AI algorithms can learn, which patients benefit most from specific imaging modalities, thus improving efficiency and efficacy of patient selection and referral. Analysing large databases can then be used to facilitate the detection of anomalies, as shown with electrocardiograms to diagnose heart failure. A bottleneck for direct learning from a collection of imaging data is the need for annotation by experts, which is limited due the lack of resources and expertise.⁶ To address this, AI could also learn itself the relevant imaging features related with a given pathology, by using temporal follow-up and outcome data, which might be more easily available. When sufficient information can be used, AI tools can learn and predict disease evolution and patient prognosis so that it can become a crucial aid for therapy selection, planning, guidance, and follow-up.⁶

Although AI has clearly caused paradigm shifts in cardiovascular imaging, which is not surprising given that DL was mainly developed for (photographic) image interpretation, the application of AI for other cardiology applications is still in its early stages. For use in interventional cardiology (IC), two intelligence axes can be considered, a virtual and a physical. The virtual axis requires cognitive computing for health management systems (i.e. electronic health records and medical image analysis software), and automated clinical decision support.⁸ The physical axis is best represented by robotic interventional approaches. AI offers the possibility of detecting patients with high-risk profiles and gauge treatment effects according to various factors. In the catherization laboratory, AI can support procedural guidance for angiography, intravascular imaging, and provide any form of additional provision to the operator during the procedure.⁸ A specific example of such an application is automated pressure waveform analysis and real-time accurate identification of damping during coronary angiography.⁹ Soon, AI could offer patient-specific, vessel-specific, or even lesion-specific revascularization strategies.⁸

Also, AI in combination with the fast implementation of mobile health (mHealth) technologies (such as smart wearables), accelerated by the COVID-19 pandemic, have delivered opportunities in screening and monitoring of cardiac health at a population wide level, as well as giving subject-specific data.¹⁰ The key applications include screening (e.g. using smart wearables to screen for atrial fibrillation); supporting diagnosis (e.g. identifying ST-segment elevation in patients with chest pain through a smartphone application); improving the analysis of imaging modalities such as computed tomography coronary angiograms and prognostication.¹

Although the AI rise seems unavoidable and can potentially result in better healthcare delivery, the appropriate integration into clinical workflows remains challenging.¹ Biases in training data, model overfitting, inadequate statistical correction, and limited transparency around the processes by which algorithms reach their output (‘black box’ systems), are only some of the pitfalls that can have important repercussions for patients and need careful assessment by researchers, clinicians, and regulatory entities. Also, important security, privacy, and explainability issues must be resolved to achieve sufficient trust by clinicians.^1,11 AI is guaranteed to be more closely intertwined with our lives, and the field of cardiology is embracing this trend and the potential it presents!

Funding

R.A. is supported by national funds through Fundação para a Ciência e a Tecnologia, within the scope of the Cardiovascular R&D Center (UIDB/00051/2020 and UIDP/00051/2020), RISE (LA/P/0053/2020) and project IMPAcT (PTDC/MED-FSL/ 31719/2017).

Authors

Bibliography: Rui is a Biologist with a PhD degree in Cardiovascular Sciences obtained in 2019 at the Faculty of Medicine of the University of Porto (Portugal), where he currently works as a postdoctoral research scientist at the Cardiovascular Research and Development Center-UnIC. Rui has a strong expertise in animal models of pulmonary arterial hypertension (e.g. monocrotaline, hypoxia-Sugen5416) and in in vivo and in vitro evaluation of cardiac function. Rui has also maintained relevant collaborations with institutions of excellence in cardiovascular research and therapeutic innovation, including INSERM (France), Medical University of Graz (Austria), Christchurch School of Medicine (New Zealand), and Antwerp University (Belgium). As an early career researcher, he has won numerous prestigious scholarships and awards such as a Janssen Innovation Award (2018) and European Respiratory Society Short-Term Fellowship Grant (2017). His current research focuses on elucidating the role and therapeutic potential of novel small molecules (e.g. small peptides and microRNAs) in the setting of pulmonary arterial hypertension and associated heart failure. He is also a core member of the Scientists of Tomorrow Nucleus of the European Society of Cardiology.

Bibliography: As Master of Engineering Sciences, Bart Bijnens obtained a PhD in Medical Sciences, after which he was appointed Associate Professor of cardiovascular imaging and dynamics at the University of Leuven, working on a combination of imaging/engineering research in cardiovascular dynamics and clinical studies on myocardial deformation. In 2005, he was at St George’s Hospital, London, supervising clinical imaging research. Since 2006, he is visiting professor at the University of Zagreb, linking Cardiology and Engineering. Since 2008, he is ICREA Research Professor in Barcelona associated to the Universitat Pompeu Fabra and IDIBAPS-Hospital Clinic, leading the group on Translational Computing in Cardiology. He is a senior translational researcher with 30 years of experience in clinical cardiology. He authored >250 papers, in Cardiac Imaging, Mechanics and Physiology, using Image Analysis, ML, and Computational Modelling. He has contributed to important developments in image-based decision making in Cardiology, including deformation assessment based on strain imaging; understanding cardiac mechanics in coronary artery disease; new concepts to assess cardiac mechanics and dyssynchrony in left bundle branch block. He was the first to use synchrotron-based X-ray phase contrast imaging to assess whole heart microstructure and is currently suggesting novel ways to use interpretable ML for clinical decision making.

References

Author notes