Artificial intelligence for blood purification in critical care

Abstract

The recent development of data science has been fundamental. It is even called the “fourth paradigm of science” since the dawn of human history; the first three were experimental, theoretical, and computational science. The high operation speed by affordable computers and growing multivariable datasets stored in electronic health records are indispensable in the current boom of artificial intelligence (AI) in medicine. AI was first applied to image recognition tasks. A deep learning algorithm for detecting diabetic retinopathy in retinal fundus photos became famous with its high accuracy compared to board-certified ophthalmologists. DeepMind, which was one of the subsidiaries of Google, published another famous report of AI, consisting of a recurrent neural network; it can also predict acute kidney injury 48 hours before it happened. Because blood purification in critical care routinely records multivariable, time-series digital data in daily clinical practice, AI programs, which assist clinicians in this field, would be feasible. Additionally, there are some challenges in using AI in clinical medicine, including, but not limited to, skill atrophy of medical staff, patients’ privacy concerns, and poor interpretability of the results given by AIs. It would be of great value if medical professionals could adapt to this drastic paradigm shift in science and pave the way to the ideal coexistence of human beings and AIs.