AI-Powered Navigation System for Steering POCUS in the COVID-ICU∗

Corresponding Author

This is one of the main challenges in machine learning, as an overfitted model is not generalizable to "never-seen-before" data. On the contrary, underfitting occurs when a model is too simple, which makes it inflexible in learning from the dataset (14). Also, considering DL has multiple layers and performs analysis in a nonlinear manner, training time is increased. This technology needs powerful computing processing units and a cloud-based system, which can be resource limiting (15).
Although these limitations have been dodging the scalability of POCUS, its appeal continues to expand in point-of-care markets with the further miniaturization of technology, improved ease of use, lower system cost, increased portability, and greater access to training. Its use has spread beyond the hospital setting, with use by paramedics and during evacuations for the battlefield by military medics, and even in more esoteric places such as on the international space station and Everest base camp (16). Its utility in remote areas with limitations to cardiac imaging expertise has been well incorporated in telehealth models. This interpretation of images has been tested on smartphone apps, making transfer and accessibility of data more feasible (17). Another evolving solution has been real-time robot-assisted remote echocardiography followed by cardiologic consultation. This workflow involving POCUS has been found to significantly reduce the total diagnostic process time (18). Considering the increased utilization, the future estimated growth of POCUS is expected to exceed $3 billion globally by 2025, up from $1.3 billion in 2018 (19).
To make certain that such technology finds its roots in our everyday practice, we need to ensure that POCUS users have the requisite training, experience, and skillset for a seamless adoption. Legislations need to fast forward for creating required credentialing for AI-based technology. This will help streamline reimbursements while ensuring patient confidentiality, access, and maintenance of secure images, operator validation, and facility accreditation (20).
Although not undermining the prospects of a technology that may change the way we engage daily, we owe a mea culpa to the idea of replacement of human brainpower by "humanoid robots" very shortly. Our current limitation being the "shallowness of DL" itself. In its current form, AI-based POCUS is at best a primitive avatar of a navigation system that helps create a basic set of information to be interpreted by an expert health care professional. The current pandemic, however, has come as a shot in the arm and has set an excellent tailwind for a bright future for this technology.
Whether it is to help design an automated navigation system for blind people, provide balloon-based Internet platforms to underserved or difficult-toinnervate areas, or use neural network architectures to generate high-quality digital maps in resourcelimited countries, AI and DL can provide a shining light in personalized care in times of limited resources and constraints (21)(22)(23). With the immense possibilities, it is only a matter of time and due diligence before AI-based health care solutions are deeply integrated to become part of our daily practice, ensuring a safer environment and improved outcomes to complex disease processes.

FUNDING SUPPORT AND AUTHOR DISCLOSURES
Dr. Sengupta is a consultant for Kencor Health and Ultromics. Dr.
Raina has reported that he has no relationships relevant to the contents of this paper to disclose.